Publication Radar.

Food similarity underlies retrieval, recommendation, substitution, and other food computing tasks, yet food representations are typically evaluated by identity preservation. We ask what happens to food similarity beyond identity: how different visual representations organize foods into relational neighborhoods, what culinary relationships these neighborhoods reveal or obscure, and how consistently these structures emerge across representations. Indian cuisine provides a challenging setting in which visually similar foods may have different culinary identities, while foods within a culinary family can vary substantially in appearance. Using an expanded Khana dataset and seven pretrained visual representations spanning convolutional, transformer-based, self-supervised, and image-text aligned models, we evaluate image-to-image retrieval and image-to-dish neighborhoods, examining culinary-family structure, contrast cases, and cross-representation agreement. We also aggregate image representations to compare dish-level visual similarity across models and against recipe-derived similarity for the same dishes. Image-level neighborhoods vary substantially across representations, including similarities across culinary-family boundaries. At the dish level, aggregation reveals common visual structure despite local variation, but this structure corresponds less strongly with recipe-derived similarity. These findings show that recognition performance alone cannot characterize relational structure: food similarity depends on representation, modality, and abstraction level rather than being a fixed property of the food collection. Understanding these structures is therefore a prerequisite to combining multiple food information modalities into a coherent notion of similarity.

Latent multi-agent systems reduce inference cost by keeping intermediate reasoning and inter-agent communication in continuous representations.\ Reusable skills, however, are commonly supplied as text or stored as layer-wise Key-Value states or skill-specific model parameters, introducing repeated processing or substantial per-skill storage.\ We introduce \textbf{SkillMem}, which represents each reusable skill as a small external latent memory and makes it directly available during latent multi-agent reasoning.\ A role-conditioned cross-attention reader allows the same memory to guide different agents according to their roles and current hidden states.\ We train the skill memories and shared reader through behavioral distillation, matching the change in token predictions induced by a textual skill while keeping the backbone frozen.\ For our experiments, we obtain textual skills from paired successful and failed multi-agent trajectories, although existing skills can also be compiled directly.\ Across six benchmarks and Qwen3 models ranging from 4B to 14B parameters, SkillMem provides strong gains for smaller backbones while preserving efficient latent multi-agent inference across model scales.\ We use the same $8\times256$ memory configuration across all evaluated model scales, requiring 9.6\,KB of skill-specific storage per skill, including its retrieval embedding.\ For a library of 100 skills, SkillMem reduces total storage by more than $38\times$ compared with alternative skill representations and achieves up to a $4.23\times$ end-to-end inference speedup.\ These results show that compact procedural memories can be reused without sacrificing the efficiency of latent multi-agent reasoning.

Identifying a small molecule from its tandem mass spectrum is limited by paired data. Foundation models offer a way around this scarcity: pretrain a spectrum encoder on unlabeled spectra and a molecule encoder on a large corpus of structures, then align the two frozen representations, as in vision-language and as the Platonic Representation Hypothesis suggests. We audit this route on MassSpecGym with spectrum encoders retrained free of test-fold contamination. Alignment needs a metric relating distinct molecules, but the objectives that produce spectrum encoders never see molecular structure: we prove that some predictive objectives fix molecular identity without fixing that metric. Moreover, we find that a candidate-based objective rewards a spectrum-independent prior over the benchmark's decoys, which we isolate. What ultimately addresses data scarcity is not architectural: simulated spectra surpass frozen geometry.

Skills, as reusable external knowledge, have demonstrated substantial potential to enhance the ability of Large Language Models (LLMs) to solve complex tasks. However, existing skill-augmented generation methods typically rely on separate models for skill retrieval and skill utilization, resulting in disconnected optimization of these two processes. To address this gap, we propose Skill-Uni, a unified reinforcement learning framework that jointly optimizes skill retrieval and utilization. Specifically, we adopt a generative skill retrieval paradigm that enables a single model to both learn how to retrieve relevant skills and leverage them for task solving. To effectively train this unified framework, we introduce a two-stage training strategy consisting of warm-up instruction tuning and GRPO, allowing the model to progressively improve its skill retrieval and utilization capabilities. Extensive experiments on five skill-related benchmarks demonstrate the effectiveness of Skill-Uni, yielding average performance gains of 66.59% and 34.82% over the base Qwen3-4B and LLaMA3.1-8B models, respectively. Besides, Skill-Uni is compatible with various skill retrieval engines, demonstrating its flexibility and generalizability. Additional in-depth analyses also provide insights into the effectiveness of Skill-Uni.

Zero-shot vision-and-language navigation (VLN) requires an agent to use past observations without navigation-specific training. We study what information an 8B-scale navigation system should retain and how it should be organized for subsequent decisions. Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories. An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets. A semantic information perspective motivates the distinction between retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well. Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments, suggesting better transfer to the evaluated real-world settings. As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success. These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.

Historical experience can guide scientific coding at multiple levels, but task-level retrieval may miss implementation lessons that become relevant during execution. We introduce MUSE, a memory management framework combining multi-level memory with selective runtime injection for scientific coding agents. MUSE builds global, approach-level, and implementation-level memory from historical trajectories. At initialization, the agent receives global memory and problem-matched approach-level memory. During execution, current interactions are used to retrieve implementation-level memory. An LLM gate selects relevant experience and generates specific review advice, or skips injection. Across four benchmarks spanning numerical problem solving, kernel optimization, and scientific software repair, MUSE outperforms retrieval and memory baselines on all evaluation metrics. Ablations support the multi-level design and gated injection, while transfer experiments show that memory remains useful across kernel benchmarks. These results show that multi-level memory and selective runtime injection enable effective reuse of experience in scientific coding. Code is available at https://github.com/hr-jin/MUSE.

Reliable legal information retrieval must ground legal reasoning in authoritative statutory sources. In Arabic judicial settings, long factual narratives must be mapped to concise articles that express abstract legal rules, creating a narrative-to-legal-rule abstraction gap. Using a benchmark of 200 synthetic Saudi Arabic judicial cases and 381 articles from the Family Law and the Law of Evidence, we compare word-level and char-4 BM25, zero-shot dense encoders, per-law retrieval, dense-sparse reciprocal rank fusion (RRF), and a proposed Decompose-and-Refine (DaR) pipeline. In DaR, a large language model splits each narrative into disputed legal issues and generates, for each issue, three retrieval channels: a statute-style paraphrase, legal keywords, and a candidate taxonomy chapter. On the whole corpus, the instructed Qwen3-Embedding-8B dense encoder reaches an MRR@10 of 0.849, the best result of any non-decomposed system. Under per-law scoring, this advantage reflects Family Law only: restricted to Evidence articles, the same encoder drops to an MRR@10 of 0.087, and fusion with BM25 raises it only to 0.207. DaR changes this picture, raising Evidence MRR@10 to 0.696 and Family MRR@10 to 0.864. Without decomposition, the two laws favor different strategies; under DaR, the same retriever achieves the best MRR@10 on both.

A memory can be outdated for present-day reasoning yet indispensable for reconstructing the past. Relevance-based retrieval overlooks this tension: semantic relevance, evidence-grounded validity, and query-specific eligibility are distinct. We propose CALM, a conflict-aware lifecycle memory governance framework that decouples persistent memory preservation from query-time admission. CALM converts heterogeneous records into evidence-grounded memory cells, updates validity and lifecycle states using support, conflict, and dependency signals, and admits query-appropriate evidence through training-free or learned gating informed by temporal applicability. Across LongMemEval-S and LoCoMo with Qwen and GPT backbones, CALM outperforms state-of-the-art baselines, achieving the best overall accuracy. Compared with simple filtering and reranking, query-conditioned admission further improves evidence selection. Conflict-blocking and robustness analyses show CALM suppresses erroneous retrieved memories while retaining useful evidence. Historical-reconstruction experiments demonstrate that preserving superseded records supports accurate earlier-state reasoning without compromising current-state answers.

Linear attention and state space models (SSMs) offer efficient alternatives to softmax attention by maintaining a fixed-size recurrent state, but this design requires careful management to preserve useful historical associations and limit interference. The delta rule helps reduce interference through correction of the value associated with each key, but it does not explicitly account for how related value features are distributed across the state. We observe that a state $\boldsymbol{S}\in\mathbb{R}^{d_k\times d_v}$ can be viewed as a bank of $d_k$ addresses storing $d_v$-dimensional values. While a key-side read $\boldsymbol{S}^\top\boldsymbol{k}$ retrieves the value associated with $\boldsymbol{k}$, a value-side read $\boldsymbol{S}\boldsymbol{e}$ aggregates addresses whose stored value features match the selector $\boldsymbol{e}$. Based on this observation, we propose Dual DeltaNet (DDN), which extends Kimi Delta Attention (KDA) with a two-sided delta rule. The key-side delta rule reads with the current key to correct its associated value, while the complementary value-side delta rule reads with selected features of the current value to reorganize the addresses it occupies. We derive a chunkwise parallel training algorithm and a rank-2 recurrence that increases arithmetic intensity during inference. Compared with KDA, GDN-2, and Mamba-3, DDN achieves the strongest overall performance across language modeling, zero-shot downstream tasks, and retrieval, while incurring modest training and inference overhead relative to KDA. Code: https://github.com/ReichtumQian/DualDeltaNet

Reliable vocal-cord localization in emergency video laryngoscopy remains visually challenging due to rapid tissue deformation, specular highlights, and instrument occlusions. We propose Local Point Tracking (LPT), a robust framework that combines a single-frame detector, local point propagation, and stateless mask refinement. LPT follows a localize first, snap second pipeline. First, a single-frame semantic detector generates sparse anchors, within which LPT initializes local query points (SAM-supported by default, with a regular grid as fallback). A point-propagation module tracks these points through detector-missed frames and recovers temporal bounding boxes from their geometry. Second, because point-derived boxes can be loose around the visible boundary, a box-snap step refines the selected box to the target boundary. This decomposition keeps temporal state in local point coordinates rather than a stateful mask or whole-region representation, while the spatial refinement module handles single-frame boundary precision without feeding mask errors back into temporal memory. Extensive evaluations on two distinct datasets demonstrate that our proposed LPT method achieves strong overall performance against existing baselines, with the best mAP_50, success AUC, and missing rate on both datasets. Furthermore, ablation studies confirm that point tracking primarily enhances temporal tracking robustness, while box-snap refinement effectively improves boundary tightness.

Multi-agent debate systems require agents to retain and revisit arguments across multiple rounds, yet limited context windows and unstructured memories make relevant information increasingly difficult to recover. Existing flat and pairwise-graph memories also struggle to capture higher-order relations, including multi-agent agreement, conflicting evidence, and evolving consensus. We introduce HEDGE, a selective memory framework that represents debate trajectories as semantic hypergraphs. Hyperedges connect related claims across agents, while intra-hyperedge relational graphs preserve fine-grained argumentative structure. A learned memory selection module (MSM), trained through supervised learning and reinforcement learning, routes each memory candidate to STORE, BACKUP, or OMIT, retaining useful information while limiting redundancy. Across seven diverse benchmarks, HEDGE consistently outperforms strong single-agent, multi-agent, and memory baselines, improving accuracy by 5.0 points on average over the strongest multi-agent system with GPT-4o-mini. The gains reach 8.2 points on conflict-heavy tasks, while HEDGE reduces memory usage by 48% and the number of debate rounds by 34%.

A streaming estimator must compute from the evidence that fits in memory. We determine when storing an accurate mean is easier than computing it. For Gaussian mean estimation over the Euclidean unit ball in $\mathbb{R}^d$, we establish the minimax rate under squared error with $K$ persistent states, a known horizon, and a public time index: $\min\{1, K^{-2/d}+\frac{d}{n} (1+\frac{\log(ed)}{1+(\log K)/d})\}$. Unrestricted computation followed by $K$-level encoding has rate $\min\{1,d/n+K^{-2/d}\}$. At a fixed sufficiently small error, statistically optimal sample size therefore requires order $d\log d$ working bits, while order $d$ bits suffice to store the answer. The lower bound tracks repeated state compression through an entropy potential obtained by Gaussian smoothing. A matching construction combines a multiscale modular counter with workspace reused across coordinate groups. In one dimension, it attains the terminal rate $n^{-1}+K^{-2}$. Any fixed confidence requirement that the realized loss not exceed the zero baseline preserves these rates. Increasing confidence creates a further memory requirement: at fixed nontrivial risk in every fixed dimension, failure probability $e^{-γn}$ requires and permits order $\log n$ working states for sufficiently small fixed $γ>0$. In the scalar problem, a fixed terminal alphabet suffices under the same protection. These bounds separate the memory needed to represent an answer from the memory needed to accumulate evidence and validate a decision.

Colonoscopy is inherently comparative: reliable visual assessment of polyps often depends on complementary views acquired under different viewpoints and imaging conditions. Retrieving such views requires polyp tracklet representations, which recent contrastive methods learn from temporal cues but still build from manually annotated bounding boxes. We propose an annotation-free framework that constructs tracklet views directly from raw colonoscopy videos, enabling temporal self-supervised learning on full-procedure videos, including frames without visible polyps. On REAL-Colon, our method achieves state-of-the-art results on tracklet retrieval and re-identification, outperforming both self-supervised and supervised baselines. We then introduce few-shot polyp retrieval, a task closer to clinical use, where a few observations of a target polyp are used to retrieve its remaining tracklets: with only two examples, our method surpasses most competing methods given ten. Finally, we show that a modest number of unlabeled procedures suffices for competitive performance, making on-site model development more practical. Code is available at https://github.com/lparolari/freerep.

Long-horizon dialogue agents require effective memory systems to recall facts, events, and relationships from interaction histories. In this work, we propose a hybrid-memory architecture comprising episodic memories and structured relational memories and a consolidation mechanism which preserves efficiency. We conduct an empirical study across two benchmarks: LoCoMo and LongMemEval-S. Our experiments demonstrate that the two memory types are consistently complementary, combining both outperform state-of-the-art memory frameworks and achieves 90.2% accuracy on LoCoMo and 85.5% on LongMemEval-S.

Long-horizon consistency is essential for video world models, yet it becomes difficult to maintain once observed content falls outside the context window. Existing memory mechanisms either retrieve history as implicit context, which remains constrained by context capacity and entangles static and dynamic information or consolidate history into explicit global representations, which struggle to model evolving dynamics and accumulate geometric errors. To overcome these limitations, we introduce \textbf{GROME}, a novel factorized memory framework that augments video world models for long-horizon coherence across both static scenes and dynamic entities. GROME operates via three synergistic components: 1)~\emph{Static-Dynamic Factorization} decouples historical evidence into static and dynamic components, preserving static scene geometry and dynamic entity states separately; 2)~\emph{Local Memory Retrieval} selects spatially aligned static history to eliminate geometric drift alongside the latest dynamic entity states to ensure temporal continuity; 3)~\emph{Training-Free Memory Injection} seamlessly grounds the factorized memory into video world models via their native conditioning pathways. Experiments on long-horizon revisit scenarios demonstrate that GROME substantially enhances scene persistence and dynamic entity fidelity over existing methods, all while preserving the visual quality and controllability of the frozen backbone without additional training. Beyond self-rollouts, GROME readily accommodates external observations as plug-and-play memory, unlocking flexible and controllable video continuation and scene exploration.

Memory in reinforcement learning raises two problems. Attention finds the relevant item by scoring the whole store, which is costly for robots under latency and energy limits. When a retrieved memory pays off, credit for acquiring it must cross the whole sequence. We argue that both are governed by one quantity, the value of information, and propose to estimate it from archived rollouts with a learned world model. We obtain that, when read at the observation, it yields a provable lower bound on the optimal value of the acquiring action, which floors its TD target and attributes this floor to the observation that earned it. Read at the decision, it defines the optimal retrieval set. A pilot retrieval system, to be trained and improved with this signal, solves a memory task with reduced context.

Vision--language models (VLMs) are tested for fine-grained understanding with image--text matching probes, a model must prefer the correct image or caption over handcrafted negative. In retrieval, the correct item must beat thousands of alternatives. We measure how these two evaluation methods relate by using two probing benchmarks, \svo{} and SugarCrepe, for retrieval tasks. For 100 image and caption queries we judge the top-10 retrievals from full dataset of six VLMs, with human annotators and an LLM assessor. We than compare how well models rank pool of items on the same queries, while changing only where the negatives comes from: the benchmark, a random draw, or the judged-incorrect retrievals of other models. On \svo{} the accuracy of retrieving a positive over handcrafted negatives is significantly higher than for retrieved ones (the correct item is preferred in 85\% against 77\% of the time). Using pool of ten retrieved negatives leads to same accuracy as when one uses pool of 200 to 430 random negatives. On SugarCrepe the order is reversed: its handcrafted negatives are harder than the retrieved ones. Incomplete relevance labels explain most of the low retrieval scores on \svo{}, where the top-ranked item is correct for 54--90\% of the queries under human labels against 6--21\% under the benchmark's labels. However, we detect no association between passing the probe for a query and retrieving a correct item first. The LLM assessor o3 agrees with individual annotators ($\kappa$ 0.57--0.61) slightly less than they agree with each other (0.64--0.68).

Retail sales forecasting underpins inventory planning, yet promotions and calendar events reshape short-term demand. Although each factor usually has historical occurrences, the complete factor combination of a target week often has few precedents. The challenge is therefore to retrieve informative partial matches and weight them by their reliability, which varies across series. We propose FARE(Factor-Aware Retrieval with Error-Guided Fusion), which ranks historical weeks separately for each known target-week factor and merges the selected weeks into a factor reference under a fixed budget. Guided by shrunk errors from completed development weeks and the current disagreement among candidates, a first gate mixes this factor reference with a shape reference from similar sales histories, and a second gate controls how strongly the mixture corrects a backbone forecast. Both gates remain fixed during testing. On Favorita, M5, Rossmann, and Rohlik, FARE reduces the weighted absolute percentage error (WAPE) of five backbones by 11.46%, 4.06%, 14.63%, and 10.16% on average. With Chronos-2 frozen, FARE also outperforms conventional and existing enhancement methods, including supervised residual correction and retrieval augmentation. Code is available at https://github.com/HousonGu/FARE .

Interactive world models generate visual content conditioned on user actions, yet maintaining spatial consistency over long-horizon exploration in real time remains challenging. Existing approaches rely on either temporal context or explicit 3D memory, both of which scale poorly because memory grows with the length of the generated stream. Moreover, explicit 3D memories condition latent video generators by re-encoding RGB renderings, which precludes real-time generation. In this paper, we introduce a Latent 3D Gaussian Memory (L3G) that combines the efficiency of feed-forward reconstruction with a compact, rapidly renderable scene representation. Building on this memory, we propose L3G-World, in which a memory updater and a video diffusion transformer (DiT) alternate autoregressively to generate spatially consistent worlds. The memory updater spawns Gaussians in newly observed regions and refines existing ones, so that the memory grows with scene extent rather than stream length. Each Gaussian further stores a VAE latent feature, allowing the memory to be rendered directly into the DiT's latent space without VAE re-encoding. Experiments show that L3G-World achieves state-of-the-art generation quality and long-horizon spatial consistency, with up to $38.6\times$ fewer 3D primitives, $3.2\times$ faster memory updates, and $359\times$ faster memory rendering relative to the most efficient prior 3D-memory baseline for each metric, enabling real-time world generation at 9 FPS.

Proactive assistants offer short, memory-grounded hints during a conversation, deciding when to assist and what to suggest. Current systems split this between a gate and a larger generator; we evaluate both together on 28,058 decision points. Stage-wise generation scores are oracle-gate ceilings: behind a real gate, end-to-end exact match is 0.2619 against a ceiling of 0.2971, and end-to-end precision is 0.2387, leaving substantial headroom. A gate pays off through precision and compute, whose saving falls from 4.39x with an 8B generator to 1.38x with a 1B one. The memory profile drives hint quality: removing it costs 24.7 exact-match points (rescaled BERTScore 0.57 to 0.35), while retrieval and a larger generator add nothing on exact match. One 1B decoder can do both, gating from an intermediate layer and generating in the same pass. At its final checkpoint it matches or exceeds the two-model pipelines on automatic end-to-end measures with 1.31-1.76x less wall-clock time than the matched 1B pipeline; across three seeds, its deeper exit ties the separate gate. A human evaluation (1,144 ratings over 32 dialogues) supports the memory effect and rates the single model level with the matched 1B pipeline and below the 8B one. Graded metrics and LlamaPIE's native judge track human ratings more closely than exact match, which we report as a strict lower bound.

Long-context models repeatedly solve closely related retrieval problems. Attention or global retrieval can retain large stores, but each forward pass again searches for what matters. Recurrent compression avoids this search only by deciding earlier what information to preserve. Neither mechanism directly exploits that useful memories at neighbouring retrieval steps may substantially overlap. We propose an alternative in which retrieval is a persistent local traversal through an address space. We separate a memory's content from its address. Utility is defined by the information revealed by retrieving the content, while geometry determines only how cheaply that information can be reached. This leads to a cost-regularised indexing objective that preserves information value inside local neighbourhoods while charging for every memory made locally accessible. Its marginal-value form gives both attractive and exclusion pressures. Thus, complementary memories should be co-retrievable when the additional value exceeds retrieval cost, whereas redundant memories should not occupy the same retrieval set solely because they are individually useful. As a proof of principle, we test only the simplest fixed-geometry instance on MiniGrid Memory. A persistent local traversal head reaches $0.95$ mean return after $0.59$M training steps with a five-token context, compared with $0.94$M for Gaussian absolute addressing and more than $2$M for categorical absolute addressing. The experiment does not test learned geometry or repeated retrieval. Instead, it motivates the hypothesis that stateful retrieval may be easier to learn than stateless. To test the full proposed method further experimentation is necessary.

Cross-modal scene and instance retrieval aims to retrieve a corresponding 3D environment or object given query 2D images, serving as a foundational mechanism for spatial reasoning in robotics and AR/VR. Existing methods achieve this by learning alignment only at the level of whole scenes or objects, which provides no supervision for the fine-grained local structure within them. In this work, we show that supervision on a set of sparse 2D-3D correspondences, without scene- or instance-level contrastive objectives, is sufficient to train a single model for both scene- and instance-level retrieval. Specifically, we propose $\textbf{CrossWeave}$, a novel framework for cross-modal retrieval that learns sparse patch-point alignment, which alone outperforms existing methods. An optimal transport-based aggregator over the aligned features gives further gains by clustering structurally redundant regions, freeing descriptor capacity for semantically distinct local features. CrossWeave achieves state-of-the-art performance on both scene- and instance-level retrieval on ScanNet and 3RScan with a $\textit{single model}$, whereas existing methods require separate models for each task. Our results suggest that whole-scene encoding is unnecessary for cross-modal retrieval and that sparse local correspondence supervision is a more effective and flexible alternative.

Financial AI workflows that reuse investor knowledge need persistent memory that integrates observations across sources and periods without losing identity or provenance. In a knowledge graph, recurring observations may use different surface forms, fragmenting relationship histories and preventing evidence from accumulating. We introduce Fusion and Identity Registry for Memory Evolution (FIRME), a registry-anchored memory graph that reconciles incoming observations to persistent canonical relationships and retains their source, time, and context. Within periods, rule-based fusion links investor-call discussions to portfolio cash-flow signals, directly or through a fixed product-topic ontology. Across periods, canonical identities and persistent aliases direct recurring observations to the same relationship history; deterministic updates summarize reinforcement, inactivity, and reactivation. We evaluate FIRME on proprietary investor-call and cash-flow records and on S&P 500 earnings-call transcripts. Compared with a historical registry that retains canonical identities and semantic fallback but omits alias memory, FIRME increases final-period Correct Match Rate by 7.8-12.1 percentage points across four dataset-backend settings under a fixed-judge and fallback-review protocol. A blinded human audit of 200 incoming observations per setting finds Correct Match Rate gains of 9.0-16.5 points and reattachment-recall gains of 22.5-29.0 points, with higher observed merge precision in all four samples. A separate audit finds valid evidence paths for 293 of 300 outputs from full within-period fusion in the investor-call setting. Extraction-faithfulness and round-trip diagnostics assess source grounding and preservation of extractor-derived graph content.

Complex questions in retrieval-augmented generation require evidence spread over several documents, and an answer is reliable only when the retriever finds all of it. Chain embedding retrievers collect such evidence step by step, re-encoding the question together with the passages retrieved so far, and we find that they are already strong: trained only on gold histories, a small chain embedding retriever that calls no language model outperforms graph-based systems and is competitive with a search agent. This training, however, leaves a gap. At test time the retriever conditions on its own retrievals, which soon contain mistakes: by the third step, about half or more of the states it visits hold a wrong passage, and these passages lower its chance of retrieving a missing gold passage well beyond what the difficulty of the states explains. We propose on-policy state training, which trains the encoder on the states it reaches itself. At every epoch the current encoder retrieves chains for the training questions; each state it visits, including those built on its own mistakes, is labeled with the earliest gold passage still missing while the other missing gold passages are masked, so that no correct continuation is penalized; and the encoder is updated contrastively on these states mixed with teacher-forced ones, so that its training distribution follows it as it learns. The resulting retriever retrieves complete evidence more often than teacher-forced training and the systems we compare, and it transfers without further training to harder long-chain variants of HotpotQA and 2WikiMultiHopQA that we construct. On the same states, it recovers from its own mistakes far more often, and controlled comparisons attribute the gain to training on the states the encoder currently reaches.

Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.

Graph analytics generates irregular, topology-dependent memory-access streams via indirect traversals, limiting data prefetching. Accurate prediction enables timely prefetches that hide memory latency. However, ML predictors couple short histories with fixed horizons, fitting the training trace's node layout rather than learning transferable graph patterns. We propose LoCoMAP, a long-context representation-learning framework for memory-access prediction with independently controlled observation history and prediction horizon. We evaluate Connected Components (CC), PageRank (PR), and Single-Source Shortest Paths (SSSP) on six synthetic graph families, with Erdos-Renyi (ER) and Grid as random and deterministic controls. We test two interventions: same-family regeneration changes connectivity, while node shuffling disrupts layout but preserves connectivity. Across 18 graph-workload pairs, extending history from 8 to 128 accesses improves F1 by 7.8%, while extending prediction horizon from 8 to 512 improves F1 by 88.3%. For CC, regeneration is less disruptive than shuffling: without retraining, long context retains 95.4% of original F1 and outperforms short context in six families. Under full shuffle, long context improves CC F1 by 33.6% across ER, Barabasi-Albert, Copying Model, and Stochastic Block Model (SBM), with a 2.06x gain on SBM, but not on Watts-Strogatz or Grid, where near-zero delta mass shifts into heavy tails beyond the predictor's range. Long context outperforms short context on six real-world CC graphs. At full shuffle, long-context PR retains 57.6% of original F1 and averages 11.0% above short context. Both SSSP models retain about 30% of original F1 under full shuffle as distance updates access vertex data in a new order. Within these bounds, long-context predictors are robust across related graphs and layouts without retraining, enabling a generalizable graph prefetcher.

Giving embodied agents the freedom to act autonomously, initiate interactions, and understand their surroundings through experience is a step toward bringing them to life in the physical world. Realizing this vision requires action, interaction, and memory to work together, so that each encounter can inform current decisions and contribute experience for future tasks. We present EvoLife, a unified navigation harness that enables a multimodal embodied agent to choose how to move, when to ask, and what experience to carry forward. Its hybrid action space supports waypoint navigation, local discrete adjustments, pixel-based navigation, and object-directed navigation. Active interaction allows the agent to seek clarification or assistance when it encounters ambiguity or difficulty, while persistent spatial memory binds observations and dialogue to places, retaining useful experience to guide subsequent visits. We evaluate eight benchmarks across four task families: vision-language navigation, object navigation, interactive instance navigation, and continuous multi-task navigation. With an average of only 1.87 active interactions per episode, our method achieves a first-visit success rate (SR) of 100% across four 100-episode sets spanning vision-language navigation and object navigation. Complete-sequence SR reaches 98% on 50 LHPR-VLN sequences and 94% on 50 GOAT-Bench sequences. In the R2R revisit setting, success weighted by path length (SPL) increases from 82.23% to 93.31% with retained experience. Multi-day physical navigation further demonstrates exploration, proactive help-seeking, and revisiting remembered places. Together, these results advance the vision of bringing embodied agents to life through autonomous action, proactive interaction, and the accumulation and reuse of experience.

Multimodal videos contain rich temporal cues that can facilitate faithful fusion. However, existing video fusion methods mainly rely on short-range temporal interactions, leaving valuable long-term information underexplored. Moreover, inconsistent fusion representations can cause flickering artifacts, undermining temporal coherence. To address these coupled challenges, we propose MemoryFusion, a two-stage framework integrating cross-temporal memory learning with progressive temporal refinement. Considering the complementary roles of adjacent and distant frames, the first stage combines a Short-Term Memory Bank for local temporal guidance with a Long-Term Memory Bank that preserves reliable historical information through a fusion-oriented update strategy. To mitigate temporal fluctuations, a Temporal Smoothing Module leverages soft motion cues for coarse background stabilization. Building upon the resulting fusion prior, The second stage further employs a lightweight 3D residual Temporal Refinement Module to correct remaining temporal inconsistencies while enhancing structural details. Extensive experiments demonstrate that MemoryFusion achieves superior fusion quality and temporal consistency over existing frame-based and video-based fusion methods. The source code is publicly available at https://github.com/Meiqi-Gong/MemoryFusion.

An effective agent memory system should adapt to a user's history and task requirements. Human-annotated supervision can guide this adaptation by providing feedback for evaluating and improving memory designs. Yet suitable labels are rarely available for new user histories, and manually constructing QA pairs and supporting evidence is costly. We introduce MemSteer, which turns unlabeled user histories into question-answer (QA) supervision with supporting evidence for memory-design optimization. MemSteer constructs an evidence-grounded answer plan that specifies how facts across related records jointly support a reference answer. The plan guides question generation, while source-grounded auditing and blind answer recovery check the resulting annotations. To study the utility of this supervision, we develop MemSteer-Opt, which combines QA-based design evaluation with evidence-guided failure diagnosis to revise memory store, retrieval, and context construction without updating model parameters. Experiments on ATM-Bench and RHELM show that this supervision is competitive with official-QA-guided optimization. Relative to Base-BM25, MemSteer-Opt improves ATM-Bench Question Score by 13.4 points and RHELM accuracy by 4.1 and 6.6 points without and with external data, respectively. The supervision also benefits two existing optimizers on ATM-Bench, while new-user experiments demonstrate continued design optimization using supervision generated from the target histories.

Test-time compute mechanisms, such as retries, confidence-based escalation and non-local routing, are conventionally deployed as black-box additions to trained models. They typically add forward passes or halting heads, but give no structural account of what the extra compute buys or what it costs against the model's stability guarantees. We therefore propose a theoretical and empirical framework for test-time compute on a Latent Dynamics Model used as the memory. When an encoder's latent state $(q,p)$ is advanced by a damped symplectic (velocity-Verlet) step of a learned Hamiltonian, every mechanism that expands access has an exact phase-volume Jacobian, a bounded energy change and an exact latch-transport law. We split reachability failures into escape failures and reach failures. A Lorentz squeeze resolves escape failures by injecting bounded, re-absorbable energy. A gated canonical translation (a one-step channel across $C_T$) resolves reach failures at $\detJ=1$ exactly. We verify this certificate stack on a designed analytic testbed. Reaching just beyond $C_T$ with the squeeze requires energy that grows quadratically in the excess distance. The gated channel's energy cost equals the potential difference between its endpoints, which is zero on the matched channel measured here, and so does not grow with distance in this testbed. An untrained state-replacing map annihilates phase volume and erases stored charge, while the gated channel transports it. On a learned physics-structured memory, we show calibrated compute rationing and evaluate the scaling limits of non-local routing against a physics-free baseline. The main contribution is the formalization and verification of the certificate stack, not benchmark performance.

Deploying LLM agents in specialized domains often require fine-grained task-specific procedures and knowledge a general model lacks, yet parameter training is costly and non-interpretable, and hand-engineered agent systems shift the cost to expert labor. Episodic memory updates at context level, but single agent adaptation struggles to alleviate cross-task interference, especially with complex tasks such as multi-application navigation or data pool analysis. Although multi-agent evolution alleviates this problem, existing methods lack clear definitions on update rules and rely on the backbone model to decide where and how much the system changes. This ill-scoped evolution makes learning follow the backbone's structural preferences rather than the observed failures: one backbone accumulates skills while another spawns agents. We present EMMA (structured Episodic Memory for Multi-Agent workflow synthesis), which treats the configuration of a multi-agent system (controller routing memory, specialists inheriting a generalist prompt, and per-step procedural skills) as episodic memory. EMMA makes the ownership and admissibility of updates explicit: a reflector assigns each diagnosed failure to the scope owning its correction and proposes one edit within it, and backbone-independent guards keep every accepted edit scoped, incremental and bounded. Across KramaBench, AppWorld and BBEH with five backbones and four baselines (ACE, GEPA, EvoMAS and ADAS), EMMA attains the best overall performance on every benchmark, up to 32.5% over the strongest baseline, maintains consistent improvements across backbones, and uses 17-38% fewer training executions than the most efficient baseline.

Memory-agent benchmarks often collapse retrieval and answer generation into one score, making it hard to tell whether a failure comes from missing an update or from not using an update that was already retrieved. We examine this distinction on the single-hop FactConsolidation task of MemoryAgentBench. With Qwen2.5-7B-Instruct on 6K-history BM25 accuracy drops from 45% to 32% when the retrieved evidence budget grows from 64 to 512 tokens, even though a gold-assisted post-hoc audit finds the newest statement in every retrieved set. Most of the errors occur on questions that also contain an obsolete statement version. Under a frozen prompt and equal four-statement evidence size, replacing one unrelated statement with the obsolete competing statement reduces accuracy from 100% to 44.6% for Qwen2.5 and from 100% to 87.8% for Qwen3-14B, with all 50 harmful paired flips classified as stale. A 32K-history replication likewise favors BM25-64 over BM25-512 (69% versus 58%). These results suggest that retrieving the correct fact is not enough. An obsolete competing statement can be more harmful than an unrelated piece of evidence of the same size.

When an LLM group's discussion is compressed into a short memory for a final decision maker, what should the memory keep? Two natural heuristics keep the rarest claims or the claims relevant to the decision criteria. On hidden-profile tasks both look attractive, because how widely a fact is held is confounded with whether it helps. We introduce CrossProfile, a diagnostic benchmark of 60 templates with automatically verified answer keys, in which the ownership of supporting, decoy, and irrelevant facts is crossed while their text is held fixed (480 tasks). Across Qwen2.5-7B, Llama-3.1-8B, and two DeepSeek models, both heuristics fail in predictable places. Selecting by relevance readmits decoys, because decoys are on topic: even a perfect relevance filter loses 23 to 31 points to rarest-first selection in the classic hidden-profile configuration. Rarest-first selection is no better than random selection on average and wins only where the rare facts happen to be the useful ones. In both cases accuracy follows the balance of evidence the memory keeps, not the policy that kept it. The decision model adds a third failure, a form of social proof: it gives extra weight to facts attributed to many agents. Removing the attribution tags, or reassigning them at random with line order fixed, raises accuracy in the classic configuration by 13 to 24 points. In natural discussions the evidence is lost before any memory acts: the complete transcript yields lower accuracy than no notes at all.

Digital SRAM compute-in-memory (CIM) macros enable energy-efficient CNN inference by eliminating off-chip weight transfers; however, peripheral control logic introduces three inefficiencies: bit- slice over-activation at reduced precision (L1), unnecessary accumula- tion depth (L2), and idle precision-transition stalls (L3). To address these, we present four synthesizable peripheral blocks: a Flush-Merge Enable (FME) controller, a Confidence-Based Early-Exit (EE) module, a Precision Register with Decoder (PR/PD), and a Bit-Serial Cycle Controller (CC), co-designed to minimize switching activity, wordline overhead, and leakage. Implemented in TSMC 65-nm CMOS at 335 MHz, the peripheral framework occupies 1769.760 μm2 post-synthesis and 1982.131 μm2 post-PnR. Gate-level VCD analysis reports 6.0706 μW leakage and 288.2902 μW (post-synthesis) / 322.8850 μW (post-PnR) dynamic power, totaling 328.9556 μW (0.2402% of system power). The EE module dominates dynamic power (58.2%), while WLG dominates area (58.1%). System evaluation on MobileNetV2 (CIFAR-100) shows efficiency scaling from 31.35 TOPS/W (8b/8b, 1.2 V) to 454.91 TOPS/W (2b/8b, 0.8 V), achieving 13.33x improvement without bitcell modification or retraining. The framework targets CNN workloads exploiting sparsity, with Transformer extension as future work.

Long-term memory enables Large Language Models (LLMs) to maintain coherent and personalized interactions across extended conversations. However, existing memory systems face two key challenges: (1) Information Entanglement, where heterogeneous information is mixed within coarse-grained memory units; and (2) Organizational Mismatch, where memory structures based primarily on semantic similarity may overlook other associations. To address these challenges, we propose ETAMem, a multi-view memory framework that decouples heterogeneous conversational information and aligns memory representations and organization with its distinct characteristics. Specifically, heterogeneous information parsing extracts atomic memory units and identifies affective information. View-specific memory structuring then organizes these units into three complementary hierarchical views: Event, Affect, and Time. ETAMem incrementally maintains memory structures with adaptively calibrated thresholds. For retrieval, entropy-based retrieval allocates query-specific budgets across views and performs hierarchical search within each view. Experiments on LoCoMo and LongMemEval-S demonstrate that ETAMem outperforms state-of-the-art baselines, achieving relative accuracy improvements of 12.46% and 7.96%, respectively, while maintaining competitive token efficiency. Further analyses demonstrate the complementary benefits of different memory views.

Structured compliance validation requires determining whether every obligation in a predefined requirement space is satisfied, violated, or unaddressed by a given document. This can be described as a coverage problem that existing retrieval and classification systems, by treating requirements as independent, are not designed to solve. We introduce \textbf{\ac{UCVE}}, a domain-agnostic framework that represents compliance requirement spaces as a \ac{CRG} and performs retrieval via spectral reranking over its reinforcement learning-weighted Laplacian. Edges between requirement nodes are learned via reinforcement learning with a traversal reward grounded in violation detection, reshaping the graph's community structure such that low-pass spectral filtering distinguishes relevant from irrelevant obligation clusters. Evaluating across three compliance domains, \ac{UCVE} averages 4.3\% matched-$K$ recall lift over flat cosine retrieval and consistently outperforms strong retrieval and graph-based baselines.

Sequential recommendation predicts a user's next interaction from historical behavior. Recent LLM-based methods recast this task as generative recommendation and increasingly adopt reinforcement learning such as GRPO for post-training. However, these methods suffer from two limitations, namely credit-blind optimization, where a sequence-level reward is shared across all generated tokens, and collaborative-blind supervision, where each training signal is derived from a single user and ignores behavioral patterns shared across similar users. To address both issues, we propose Retrieval-Privileged on-policy Self-Distillation (RPSD). For credit-blind optimization, we introduce on-policy self-distillation (OPSD), in which a teacher with privileged knowledge re-evaluates the student's on-policy response, and the token-level confidence difference between the student and teacher reshapes the shared sequence-level advantage into token-specific credit. However, the privileged knowledge in vanilla OPSD is still derived from a single user, leaving collaborative-blind supervision unresolved. We therefore ground the teacher in collaborative behavioral evidence from similar users. Specifically, we use temporally aware optimal transport (OT) to compare interaction trajectories across users by jointly considering item similarity and temporal structure, and further apply diversity-aware selection to retain complementary trajectories as privileged evidence. In this way, on-policy self-distillation provides fine-grained token-level credit, while OT enriches the teacher with structured cross-user evidence. To the best of our knowledge, this is the first work to bring on-policy self-distillation to sequential recommendation and to use cross-user behavior as its privileged signal. Extensive experiments across multiple real-world datasets demonstrate that RPSD achieves state-of-the-art performance.

Vision-Language-Action (VLA) policies demonstrate remarkable semantic generalization, yet remain highly vulnerable to out-of-distribution (OOD) visual shifts during execution. While existing test-time methods attempt to suppress spurious visual cues, they fundamentally overlook the rich, uncorrupted kinematic priors embedded in the policy's own training experience. We introduce Retrieve to Act (R2A), a test-time scaling framework that formulates OOD robustness purely as a geometric motion primitive retrieval problem. R2A bypasses fragile visual features by structuring successful training demonstrations into an offline spatiotemporal motion primitive graph. During inference, a query-dependent Graph Neural Network dynamically localizes the robot's ongoing execution within this graph, retrieving structurally compatible future motion primitives. These retrieved geometric priors are then seamlessly harmonized with the frozen VLA's predicted actions via a learned, dimension-wise gating mechanism, adaptively rectifying erroneous spatial trajectories while preserving the policy's local reactivity. Extensive experiments across three state-of-the-art VLA architectures, GR00T N1.7, $π_{0.5}$, and StableVLA, demonstrate that R2A consistently and significantly enhances robustness against diverse contextual perturbations. R2A achieves absolute success rate improvements of up to +12% in simulation benchmarks (LIBERO-Plus, RoboTwin2.0-Plus) and +40% in real-world deployments, without requiring any finetuning of the base policies. Project page: https://retrieve2act.github.io/r2a/

Humanoids and other legged robots operating in complex environments must remember information that leaves their field of view and only becomes relevant much later. Reinforcement learning policies address this using memory networks such as LSTMs that are trained end-to-end with the policy to memorize such task-relevant aspects. Yet, such architectures often fail to preserve sufficient information over long horizons. In this work, we propose a network architecture based on Hadamard memory that specifically targets robot navigation. Our Local Spatial Matrix Memory (LSMM) enables RL agents to store generic task-relevant information at specific, position-based memory cells, enabling flexible and spatially grounded memorization. We evaluate LSMM on the task of autonomous maze exploration, where it matches the maze coverage of strong recurrent and finite-history baselines while being the only architecture that remembers the maze layout, reducing the reconstruction error threefold.

Coverage-evaluation benchmarks quantify whether retrieved evidence covers the different pieces of information needed to satisfy an information need. Existing benchmarks mostly consider factual claims that may already be known to pretrained retrieval components, leaving open whether retrievers find corpus evidence when it contradicts those priors. We study how retrievers behave when their parametric knowledge and the information in a corpus disagree. To do so, we develop a process to augment a nugget-based coverage benchmark with controlled contradictions. For each nugget, we create a negated counterpart and edit its supporting documents to support that negation while preserving the context. Applying this process to CoverageBench, a benchmark measuring information coverage, we create CoverageBench-Negate, spanning six datasets and roughly 63,000 original-edited evidence pairs (one per nugget-document link). We attach rich construction metadata to every pair and additionally provide out-of-family LLM judgments over all pairs and a human audit of a sample. In baseline experiments, BM25 separates original from edited near-twin evidence substantially better than dense retrievers (Qwen3 and Jina), which stay near chance. We release the benchmark on Hugging Face Datasets (nuggets, documents, pairs) together with the construction materials, including the prompts used to create it.

A key challenge of RAG systems is retrieving information from enterprise documents created by multiple teams over time, where it is not possible to enrich the information using the latest information and metadata available outside the scope of organizational documentation responsibility. We investigated whether using a knowledge graph to store metadata and frequently changing information would improve upon knowledge retrieval over similarity search over embeddings. To perform the experiment, we created a snapshot of the SEC EDGAR dataset for a quarter, comprising 167,050 filings from 10,511 companies and 508,714 embedded chunks of risk-factor text. To simulate the enrichment, we created a knowledge graph of 430,257 nodes and 561,144 edges carrying an accession number on every edge. To evaluate the gain obtained via a graph-based design, we evaluated three architectures: text-to-SQL over typed columns, dense-vector RAG with HNSW cosine search, and a two-stage GraphRAG design. All architectures share the same model and temperature to allow comparison. Results are scored programmatically by an LLM as judge over 20 questions divided across four categories, each question asked three times. GraphRAG solved 16 of 20 questions on all three attempts, compared to 10 for text-to-SQL and 3 for vector RAG. GraphRAG lost on structured aggregation queries. Vector RAG lost the single-document category it was expected to win, passing 3 of 15 where GraphRAG passed 15 of 15, because company attribution is not recoverable from cosine similarity at this scale. All six confident falsehoods in the benchmark were text-to-SQL conflating distinct people who share a surname, the failure that resolving identity to an identifier before reading any text prevents structurally; GraphRAG made none. We claim no advantage beyond this corpus, this model and this question set.

Embedding-level scaling expands the capacity of language models through lookup-based memory rather than by increasing model depth, width, or dense computation. Extending this idea to vision-language models (VLMs) is challenging because visual features are continuous, spatially structured, and query-dependent, unlike discrete text tokens that naturally support n-gram lookup. We introduce V-LUMEN, a visual lookup memory for VLMs. V-LUMEN discretizes visual features with a frozen codebook, builds spatially aware memory keys through spatial aggregation-based hashing, and modulates them with text-conditioned relevance signals; all addressing is non-parametric, and the retrieved embeddings are fused into the backbone through a gated residual pathway. On five reasoning-intensive benchmarks, V-LUMEN substantially improves a frozen Qwen3-VL-2B backbone and transfers zero-shot to three benchmarks whose images are disjoint from the training data. At a matched parameter budget, moving part of a LoRA or MoE adapter's capacity into V-LUMEN preserves accuracy on two backbones, indicating that external visual memory is an independent capacity axis rather than a stronger adaptation method. Because lookup cost is constant in table size and incurred only once at prefill, this capacity grows in storage rather than in per-token compute. Code is available at https://github.com/miso-choi/V-Lumen.

Retrieval-augmented generation (RAG) systems ground large language models (LLMs) by retrieving external evidence and conditioning generation on the retrieved context. Although dense retrieval is effective for local semantic matching, independently ranked retrieval can return redundant passages, miss cross-chunk dependencies, and provide incomplete evidence for multi-hop reasoning. We propose SpectRAG, a spectral graph-augmented retrieval framework that combines hybrid candidate seeding, weighted chunk graph construction, graph-based expansion, spectral relevance propagation, and token-budgeted context selection. Document chunks are represented by dense semantic embeddings and graph-spectral coordinates derived from a weighted chunk graph whose edges encode semantic similarity, document adjacency, section hierarchy, shared entities, citation links, and similarity between LLM-generated utility questions. At query time, SpectRAG builds a high-recall seed set using exact inner-product search and hybrid relevance signals, expands it over the chunk graph, re-scores candidates using graph proximity and spectral information, and assembles the final context under a fixed token budget. The goal is not merely to retrieve individually similar chunks, but to construct a context that is relevant, diverse, structurally connected, and sufficient for grounded reasoning. Experiments on MultiHopRAG, UltraDomain, and long-document question-answering settings show that SpectRAG improves broad evidence recovery and achieves competitive generation quality against dense, hybrid, HyDE, LightRAG, and NaiveRAG baselines. The results further indicate that graph-spectral structure is particularly useful for inference-oriented and cross-document retrieval, while highlighting trade-offs in early precision, computational cost, and domain-dependent performance.

Proactive LLM agents anticipate service opportunities from ongoing observations and accumulated history. Existing approaches emphasize recent context or retrieval of related memories, providing limited support for validating recurring behavioral regularities and assessing their applicability to the current event. We propose Proactive Memory (ProactMem), which organizes history into recent events, temporal trajectories, and behavioral regularities. Trajectory facts are checked against source events, and regularities are assessed through historical replay before being matched to current conditions. We also construct ProactiveStream, a longitudinal benchmark covering three application scopes, where agents process isolated event streams chronologically under episode-aware opportunity labels. On ProactiveStream, ProactMem achieves an overall effectiveness (F1) of 52.6%, a 21.4% relative improvement over the strongest of five external baselines. Code is available at: https://anonymous.4open.science/r/ProactMem-5351/.

Generative retrieval models have been increasingly adopted in search and recommendation systems. However, industrial generative retrieval models are commonly specialized to a fixed online scenario: they learn from scenario-specific interactions, support a narrow set of query forms and retrieval objectives. We study how to remove these couplings and build a unified end-to-end multimodal search system for open-ended image--text queries, heterogeneous shopping intents, and online scenarios with different resource constraints. We present \textbf{PailitaoMMSearch}, which directly generates product Semantic IDs from a raw query image and free-form instruction and scales along three dimensions. \textbf{Data scaling} uses a reusable toolset that can be composed into intent- and lifecycle-specific workflows for cold-start synthesis, quality control, real-traffic refinement, and continuous data production. \textbf{Intent scaling} progressively equips one model with e-commerce product knowledge, multiple retrieval relations, and adjustable reasoning budgets through continual pre-training (CPT), mixed-budget supervised fine-tuning (SFT), and unified reinforcement learning (RL). The RL stage combines intent-balanced optimization and Sample--Beam rollouts aligned with multi-candidate serving. \textbf{Scenario scaling} exposes quality--cost operating points across model capacity and reasoning budget. The system is deployed in multimodal search applications on Taobao Pailitao. A controlled online A/B test reports statistically significant improvements, including item page views $\textbf{+17.90\%}$, transactions $\textbf{+27.04\%}$, and gross merchandise value $\textbf{+8.94\%}$.

A large language model maintains conversational content in an external storage that is outside the model and grows as the dialogue continues. This work assumes an environment where a language model has no external memory and no prior knowledge of the future question or its answer. The motivation is to research whether a dialogue can be stored in the model parameters and later answer a question about it. This work proposes self-keyed in-parameter memorization (SKIM), a closed-form method in which the model generates its own questions about each dialogue turn and writes those questions into its parameters as the keys that later retrieve the turn's content. Generating many varied questions for each turn allows a later user question to match one even when phrased differently, and its answer is recalled from the parameters alone. Evaluated on five language models, SKIM recalls most stored dialogue turns from later user questions, and it answers from the model parameters without any external storage.

Large language model (LLM) agents rely on memory systems to leverage long-term interactions, yet existing systems typically retrieve memories based on their relevance to the current query and use them as contextual evidence. However, relevance alone does not characterize how a memory contributes to task solving: the same memory may serve different roles across tasks, such as directly supporting the answer or linking to complementary memories with weaker semantic relevance but necessary information. We therefore reframe agent memory systems as Task-Oriented Role-Adaptive Workspaces, where memories are selected and organized by their task-oriented roles and relations. Based on this view, we propose TrawMem, which organizes interaction histories into local threads and constructs task-oriented workspaces by assigning roles and jointly selecting complementary memories. Experiments on LoCoMo and MemGallery with four open- and closed-source backbones show consistent improvements over representative memory systems, with macro-F1 gains of 2.23-3.14 points. On LoCoMo, TrawMem improves the LLM-Judge score over SimpleMem by 4.07 and 4.31 percentage points on Qwen3.5-9B and DeepSeek-V4-Flash, respectively, while requiring approximately 12.8% and 5.1% fewer context tokens, respectively.

Sparse Mixture-of-Experts (MoE) models offer scalable capacity for Web applications, but uneven routing creates computation and communication bottlenecks. Existing placement and replication methods redistribute this workload, yet additional expert copies compete with token-related buffers for limited GPU memory. We present LRMoE, a memory-aware MoE training system that jointly recalibrates backward demand and balances the retained workload. During forward execution, it plans retention quotas and expert placement under resource constraints while preserving all routed computations. During backpropagation, a load recalibration metric estimates each token-expert pair's contribution to the expert update and selects pairs within these quotas. A capacity-aware routing algorithm distributes retained pairs among eligible instances of their original experts. Physical compaction then forms smaller communication payloads and variable-size grouped GEMMs, reducing backward traffic, computation, and execution-buffer demand. Experiments on Qwen3-30B-A3B and GLM4.5-106B-A12B show that LRMoE reduces cumulative MoE-layer training time to fixed-loss targets by 21.2-44.8% relative to SYMI.

AI systems that ground their answers from retrieved documents usually need only a few relevant passages, but applications such as systematic reviews and legal discovery need nearly all of them, because their conclusions rest on the full body of evidence. These applications have traditionally searched with curated taxonomies, which can only find what their vocabulary describes. We compare taxonomy-based retrieval with an alternative neural retrieval based on sentence embeddings and followed by language model screening in ImpactAI, an evidence synthesis application built on 7,354 randomized trials. Measured against the studies included in 29 systematic reviews, the share of studies found rises from 45% to 85%, and the screening model agrees with human annotators as well as annotators agree with each other. This gain comes at a cost in speed: at the median, retrieving and screening evidence for a query takes 77 seconds, against 6 seconds for taxonomy retrieval. The results suggest finding evidence with embeddings, organizing it with taxonomy labels, and setting how many candidates to screen according to the time available for each query.

Instruction-following information retrieval (IFIR) requires retrievers to rank documents by both a query and its instance-specific instructions, yet instruction-conditioned contrastive supervision remains scarce. Existing synthetic methods typically fix a positive passage and generate additional negatives, overlooking the hard-negative potential of the original positive. We propose Natural Hard-Negative Mining (NHNM), which promotes convertible negatives under query-compatible instructions and repurposes original positives as natural instruction hard negatives. Its automated Candidate Cleaning, Reversal Construction, and Geometry Filtering pipeline produces compact, high-contrast supervision aligned with the pretrained retrieval space. Applied to MSMARCO, NHNM yields NHNM-MARCO, an IFIR training set of nearly 7K instances. On InstructIR and FollowIR, models with 568M to 4B parameters fine-tuned on NHNM-MARCO achieve strong instruction-following performance and match or surpass substantially larger retrievers in several settings. Controlled analyses and ablations attribute these gains to natural hard-negative supervision rather than model scale or training-data volume.