Proposes a theory that knowledge distillation improves performance through easier optimization landscapes, not better models
Explore technology by industry and role
Newest additions first. Explore past and current notes by industry and role.
All notes
Newest additions first. Explore past and current notes by industry and role.
A novel attack method dynamically manipulates object detection by using non-co-located adversarial patches.
Quantitative assessment of spatial awareness and traffic rule adherence in LLM-based autonomous driving systems
A dynamic routing strategy selects LLMs during search to maximize heuristic quality and efficiency
LLMs generate promising low-energy parameters for structured Hamiltonians, but show larger errors for general Hamiltonians
LiveGeo benchmark introduces a method to assess agents' ability to acquire information, verify evidence validity, and revise plans in continuously changing geos
Integrates visual data with language models to enable planning in complex driving scenarios
Sensitive user information is inferred from conversations and stealthily encoded into harmless-looking model outputs
Proposes a method that optimizes policies using only verified negatives, even when positive verification is unavailable.
Extends the LSFM pedestrian detector to general object detection for real-time traffic scene analysis.
Multiple sub-tasks allow precise localization of the next sub-goal in sparse attention heads.
A closed-loop infrastructure enables continuous learning and improved generalization of multimodal LLMs in autonomous driving.
Uses explicit frame-to-frame correspondences to guide output generation and improve temporal consistency
GenAD proposes a generative framework that directly predicts future evolution of ego and surrounding vehicles from raw sensor inputs.
Integrates vision-language models to better understand complex road conditions and human behaviors.