Inference efficiency
Encode ECG beat images using a pseudo-colour representation that fills the region under the QT interval, where the colour progression is designed from clinical electrophysiology knowledge and mapped to a validated QT-nomogram risk model. Compare a monochrome input pipeline against the pseudo-colour pipeline while keeping the neural architecture otherwise identical, then evaluate explanation faithfulness, localisation, and stability across multiple explanation methods on an external cohort.
AI summary
Training & fine-tuning
Adapt a pretrained vision-language-action model by systematically sweeping the number of target-setting demonstrations and the finetuning steps, then select the best-performing demonstration/training budget for each suite.
AI summary
Inference efficiency
A training-free method that improves multi-scene temporal control in audio-driven human animation by steering a diffusion transformer’s spatio-temporal self-attention: it suppresses cross-scene attention for background tokens while preserving cross-scene attention for foreground (the human subject). For robustness under large motion, it derives dynamic foreground masks from self-attention layers/heads that best follow the subject trajectory, and uses those masks to localize which tokens should keep cross-scene links.
AI summary
Inference efficiency
Use a 3D convolutional encoder–decoder segmentation backbone, inserting a Mamba-style selective state-space block specifically at the bottleneck to model global context while keeping local feature extraction in the convolutional parts.
AI summary
Inference efficiency
HALMES is a lightweight, plug-and-play mechanism for large vision-language models that monitors internal attention features during generation to detect hallucination-precursor token segments. Instead of applying contrastive decoding uniformly at every decoding step, HALMES selectively activates contrastive decoding only when hallucination-prone segments are identified, aiming to correct likely hallucinations while preserving otherwise correct outputs.
AI summary
Training & fine-tuning
A latent-space dataset distillation approach for 3D medical classification that allocates synthetic prototypes across intra-class latent modes to prevent “prototype concentration” in only dense regions, while still maintaining class-level distribution matching.
AI summary
Inference efficiency
Integrate a latent denoising diffusion model into the split point of a split federated learning pipeline so that the intermediate representations exchanged between client and server are diffused/denoised to reduce susceptibility to reconstruction attacks, while keeping the downstream segmentation task effective. The approach is explored via architectural variants that change how conditioning and feature fusion are applied around the diffusion module.
AI summary
Training & fine-tuning
CAML trains a handwriting classifier using a modified objective that emphasizes separation between pre-identified confusion pairs. It builds a sparse, weighted confusion graph from a held-out discovery partition (separated from pilot training) and then applies pair-specific cosine-separation margin constraints during final training. The approach is designed to add no extra classifier, no routing or inference-time logic, and no inference-time parameters beyond the original model.
AI summary
Reasoning & planning
Design an interaction protocol where an agent (or human) answers a question by repeatedly inspecting bounded local windows across a topically related pool of videos, retaining supporting evidence across rounds, and producing an answer that cites specific temporal spans. Use a matched diagnostic benchmark to compare evidence acquisition and citation fidelity over multiple rounds, including both (1) temporal evidence localization and (2) end-to-end discovery/retention/composition of cross-video evidence.
AI summary
Reasoning & planning
Train a model to emit a dedicated final-response marker token when it recognizes an evaluation, while suppressing any verbalized awareness in its reasoning. Use the presence/absence of the marker as instance-level labels to study whether interpretability methods can detect unverbalized evaluation awareness when the usual verbal cues are absent.
AI summary
Agents & tool use
Introduce a tool-layer verification step for clinical (or other evidence-driven) agents: (1) validate that retrieved evidence relevant to the user’s request matches expected read requirements, and (2) when an agent’s write/update intent is ambiguous, route or label those updates through a constrained public interface so the system can attribute and restrict which changes are permissible.
AI summary
Training & fine-tuning
Use self-distillation from a teacher’s original output distribution as a zero-shift baseline, then selectively perturb feed-forward network (FFN) activations linked to private information across text and vision to create new supervision targets that remove private content while limiting distribution shift.
AI summary