AI/ML arXiv cs.AI

Ensemble Distributionally Robust Bayesian Optimisation with Continuous Context

EDRBO introduces ensemble distributionally robust Bayesian optimization to handle uncontrollable environmental contexts and distributional mismatch in continuous context spaces.

AI/ML arXiv cs.AI

When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models

Researchers identify geometric over-alignment in VLMs as a cause of hallucinations and propose a training-free inference strategy to project out linguistic bias.

AI/ML arXiv cs.AI

A Simplex Witness Certificate and Escape Force for Constant Collapse in Variational Autoencoders

This paper proposes a simplex witness certificate and escape force to prevent constant collapse in Variational Autoencoders (VAEs).

AI/ML arXiv cs.AI

Open-source LLMs administer maximum electric shocks in a Milgram-like obedience experiment

A study reveals that many open-source LLMs comply with harmful requests under sustained authority pressure, mirroring the Milgram obedience experiment.

Software Engineering Hacker News

Markdy: Like Mermaid Diagrams, but for Motion

Markdy is a tool that allows users to create motion diagrams using a syntax similar to Mermaid Diagrams.

AI/ML arXiv cs.AI

Event-Grounded Question Answering over Long Audio via Structured Retrieval

LA-RAG is a structured framework for question answering over long audio files using retrieval-augmented generation and a SQL event database.

AI/ML arXiv cs.AI

MyoInteract: A Framework for Fast Prototyping of Biomechanical HCI Tasks using Reinforcement Learning

MyoInteract is a framework for fast prototyping of biomechanical HCI tasks using reinforcement learning, reducing training times significantly.

Hardware/Chips arXiv cs.AI

Bitwise Systolic Array Architecture for Runtime-Reconfigurable Multi-precision Quantized Multiplication on Hardware Accelerators

A new bitwise systolic array architecture for hardware accelerators enables runtime-reconfigurable multi-precision quantized multiplication.

AI/ML arXiv cs.AI

No Certificate, No Categorical Speech Act: A Brouwerian Assertibility Constraint for Public Reason

The paper proposes an assertibility constraint for AI in high-stakes domains, requiring a publicly inspectable certificate of entitlement for assertions.

Other arXiv cs.AI

An Approach to Simultaneous Acquisition of Real-Time MRI Video, EEG, and Surface EMG for Articulatory, Brain, and Muscle Activity During Speech Production

Researchers developed a multimodal acquisition paradigm to simultaneously record MRI video, EEG, and surface EMG for speech production analysis.

Hardware/Chips arXiv cs.AI

CRAFT: A Tendon-Driven Hand with Hybrid Hard-Soft Compliance

CRAFT is an open-source, tendon-driven anthropomorphic hand with hybrid hard-soft compliance designed for contact-rich manipulation.

AI/ML arXiv cs.AI

AI-Driven Predictive Maintenance with Environmental Context Integration for Connected Vehicles: Simulation, Benchmarking, and Field Validation

A contextual data fusion framework for predictive maintenance in connected vehicles integrates internal sensors with external environmental data for edge inference.

AI/ML arXiv cs.AI

HiPath: Hierarchical Vision-Language Alignment for Structured Pathology Report Prediction

HiPath is a lightweight VLM framework for structured pathology report prediction, achieving high accuracy and safety in clinical settings.

AI/ML arXiv cs.AI

Policies Permitting LLM Use for Polishing Peer Reviews Are Currently Not Enforceable

A study finds that policies prohibiting LLM use for polishing peer reviews are currently unenforceable due to high misclassification rates by AI detectors.

AI/ML Hacker News

The Unbearable Cheapness of Open Weight Models

A discussion on Hacker News regarding the decreasing cost and increasing accessibility of open-weight AI models.

AI/ML arXiv cs.AI

MOCHA: Multi-modal Objects-aware Cross-arcHitecture Alignment

Introduces MOCHA, a distillation framework that transfers multimodal knowledge from a frozen VLM teacher to a lightweight vision-only detector for efficient personalized object detection.

AI/ML arXiv cs.AI

ATHENA: Agentic Team for Hierarchical Evolutionary Numerical Algorithms

Presents ATHENA, an agentic framework that emulates scientific research to autonomously design and refine complex numerical workflows and scientific pipelines.

AI/ML arXiv cs.AI

Computing Evolutionarily Stable Strategies in Imperfect-Information Games

Proposes a sound algorithm for computing evolutionarily stable strategies (ESSs) in symmetric perfect-recall extensive-form games of imperfect information.

AI/ML arXiv cs.AI

EMFusion: Uncertainty-Aware Conditional Diffusion Model for Multivariate Narrow-band Exposure Forecasting

Introduces EMFusion, a conditional diffusion-based framework for uncertainty-aware multivariate narrow-band electromagnetic field forecasting.

Cybersecurity arXiv cs.AI

Attention in Motion: Secure Platooning via Transformer-based Misbehavior Detection

Presents AIMformer, a transformer-based framework for real-time misbehavior detection in vehicular platoons to secure V2X communications.