Driving · 2026-10-28
Readout and Operating-Point Sensitivity in Event-Stream Candidate Selection
Yifan Shuai
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors evaluate offline, action-defined event-stream candidate selection on DailyDVS-200 with participant-disjoint splits and sensitivity analyses. They report that representation readout (patch mean vs CLS) and validation-threshold operating points can change model rankings: e.g., DINO-S CLS captures 101/132 with 40/388 false triggers, while patch mean captures 78/132 with 35/388. Lightweight supervised encoders are competitive and the authors find no established JEPA-specific advantage.
Why it matters
The paper shows that capture/false-trigger comparisons depend on readout choice and calibration to the operating point (threshold selection under FPR budgets), so size-based conclusions may not transfer. By using low-FPR ROC diagnostics, repeated participant repartitioning, and threshold recalibration, the authors provide evidence that representation comparisons can reverse across metrics and that lightweight baselines matter when evaluating event-candidate selection.
Method
- Offline candidate selection on a DailyDVS-200 subset with deterministic binary policies (P1: handshake or fall; P2: adds door opening), using participant-disjoint train/validation/test splits.
- Frozen encoders with patch-mean vs CLS readouts; per-fold ridge regression and validation-threshold selection under FPR budgets, then evaluation on test folds.
- Post-review extensions include additional supervised-pretrained CNN encoders (MobileNetV3-Small; ResNet-18; TinyCNN) plus repeated participant repartitions, threshold recalibration bootstraps, and operating-point sensitivity analyses.
Limitation
The evidence concerns offline action-defined candidate selection on previously used development data; it does not establish personal recording value, continuous false alarms per hour, or eyewear energy.
Abstract (from arXiv)
Event-triggered recording needs a useful combination of target capture and false triggers, yet representation comparisons can change with readout and threshold calibration. We evaluate real event accumulations from 520 DailyDVS-200 clips and 47 participants using participant-disjoint training, validation, and testing. The original study varies frozen encoder scale, accumulation duration, and temporal view count; a post-review extension adds two supervised-pretrained convolutional encoders, a 61,586-parameter trained CNN with three seeds, ten repeated participant partitions, and calibration sensitivity. At 50 ms/eight views under the primary handshake-or-fall policy, DINO-S CLS captures 101/132 targets with 40/388 false triggers, versus 78/132 and 35/388 for patch mean. DINO-B CLS has lower selected-threshold capture but higher fold-wise low-FPR partial AUC (0.746 versus 0.732). MobileNetV3-Small captures 97/132 with 53/388 false triggers, demonstrating why lightweight baselines and actual operating points matter. Across ten dependent repartitions, mean normalized partial AUC is 0.738 for DINO-B CLS and 0.674 for MobileNet. The original LeJEPA S-to-B capture difference is 8.3 points, but its multiplicity-adjusted interval includes zero. Duration and view-count effects remain policy dependent. We find no established JEPA-specific advantage. The evidence concerns offline action-defined candidate selection on previously used development data; it does not establish personal recording value, continuous false alarms per hour, or eyewear energy.
Related papers
- Trusting Imagined Futures: Auditing Imagined Rewards for VLA Post-Training
- ReWorld-Track: A Recursive Event World Model for Language-Guided Multi-Camera Tracking
- How Much Can Best-of-N Help a World Action Model? Measuring Selectable Headroom and Realizing It in Closed Loop
- VeriVLA: Adaptive Test-Time Verification for Flow-Based VLAs with Latent World Models
- Foresight at the Event Boundary: Evaluating Physical Prediction in Video World Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-27 · Indexed by Pub-AI: 2026-10-28 UTC
Readout and Operating-Point Sensitivity in Event-Stream Candidate Selection
Preprint posted: 2026-10-27 · Original record ↗
- arxiv
- 2610.36743
Explicit source versions (1)
- v1 · 2026-10-27
- Preprint available
Source date: 2026-10-27 · First detected: 2026-10-28 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2610.36743v1; arxiv:announce_type=new; item pubDate (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Event-triggered recording needs a useful combination of target capture and false triggers, yet representation comparisons can change with readout and threshold calibration. We evaluate real event accumulations from 520 DailyDVS-200 clips and 47 participants using participant-disjoint training, validation, and testing. The original study varies frozen encoder scale, accumulation duration, and temporal view count; a post-review extension adds two supervised-pretrained convolutional encoders, a 61,586-parameter trained CNN with three seeds, ten repeated participant partitions, and calibration sensitivity. At 50 ms/eight views under the primary handshake-or-fall policy, DINO-S CLS captures 101/132 targets with 40/388 false triggers, versus 78/132 and 35/388 for patch mean. DINO-B CLS has lower selected-threshold capture but higher fold-wise low-FPR partial AUC (0.746 versus 0.732). MobileNetV3-Small captures 97/132 with 53/388 false triggers, demonstrating why lightweight baselines and actual operating points matter. Across ten dependent repartitions, mean normalized partial AUC is 0.738 for DINO-B CLS and 0.674 for MobileNet. The original LeJEPA S-to-B capture difference is 8.3 points, but its multiplicity-adjusted interval includes zero. Duration and view-count effects remain policy dependent. We find no established JEPA-specific advantage. The evidence concerns offline action-defined candidate selection on previously used development data; it does not establish personal recording value, continuous false alarms per hour, or eyewear energy.
