AI/ML arXiv cs.AI

Von Mises Based Uncertainty Quantification for Closely Spaced Automotive Radar Targets

A study on uncertainty quantification for automotive radar targets using von Mises ensemble (ENS) and evidential deep learning (EDL).

AI/ML arXiv cs.AI

Robustness of Robotic Manipulation: Foundations and Frontiers

A systematic study providing formal definitions and guiding principles for achieving robustness in robotic manipulation.

AI/ML arXiv cs.AI

FinPersona-Bench: A Benchmark for Longitudinal Psychometric Stability of Autonomous Financial Agents

FinPersona-Bench is a benchmark for evaluating the psychometric stability of autonomous financial agents over long horizons.

Hardware/Chips TechCrunch

Apple is reportedly planning new iPad Pro and MacBook Pro releases early next year

Apple is reportedly preparing to release new iPad Pro and budget-friendly MacBook Pro models early next year.

AI/ML arXiv cs.AI

3D HAMSTER: Bridging Planning and Control in Hierarchical Vision Language Action Models through 3D Trajectory Guidance

Researchers introduce 3D HAMSTER, a hierarchical framework for robot manipulation that uses 3D trajectories to improve planning and control over 2D-guided models.

AI/ML arXiv cs.AI

Beyond Binary Instrument QA: Probing Instrument Grounding in Music Audio-Language Models

A new diagnostic benchmark sequence based on OpenMIC is introduced to probe whether music audio-language models truly ground instruments or rely on shortcuts.

AI/ML arXiv cs.AI

PGUDA: Pressure-Guided Unsupervised Domain Adaptation with Cross-Modal Knowledge Distillation for sEMG-Based Gesture Recognition

The PGUDA framework uses pressure-guided unsupervised domain adaptation to improve the accuracy and label efficiency of sEMG-based gesture recognition.

AI/ML arXiv cs.AI

From Materials Database to Materials Bank: Assetizing Data for AI Driven Materials Innovation

The 'Materials Bank' proposal suggests a value-filtering layer to transform materials databases into actionable assets for AI-driven innovation.

AI/ML arXiv cs.AI

Calibrating the Evaluator: Does Probability Calibration Mitigate Preference Coupling in LLM Agent Feedback Loops?

A study finds that probability calibration in LLM agent feedback loops can significantly reduce evaluator preference coupling and improve stability.

AI/ML arXiv cs.AI

Stage-Transition Dense Reward Modeling for Reinforcement Learning

Stage-Transition Dense Reward (STDR) is a framework that converts expert videos into dense rewards for more efficient reinforcement learning in robotic manipulation.

AI/ML arXiv cs.AI

Resolving superposition in AI for interpretability and cross-modal alignment in patient-neuronal images

Researchers use sparse autoencoders (SAEs) and Gromov-Wasserstein optimal transport to resolve superposition in biological images for better interpretability.

AI/ML arXiv cs.AI

Mixture-of-Control: State-Aware Fine-Tuning for Transformer-based Models

Mixture-of-Control (MoC) is a lightweight fine-tuning framework for Transformers that integrates local and global control signals to save memory and improve representation.

AI/ML arXiv cs.AI

Visual Semantic Entropy: Do Vision Language Models Recognize Visual Ambiguity?

Visual Semantic Entropy (VSE) is proposed as a new method for estimating uncertainty in Vision-Language Models by perturbing images rather than text.

Other Hacker News

Healthy but Sedentary People Show Early Decline in Cellular Energy Production

A study indicates that healthy but sedentary individuals experience an early decline in cellular energy production.

Cybersecurity Hacker News

Opening up 'Zero-Knowledge Proof' technology to promote privacy in age assurance

The use of Zero-Knowledge Proof technology is being explored to enhance privacy in age assurance processes.

Other Hacker News

How do wombats poop cubes? Scientists get to the bottom of the mystery

Scientists have researched the biological mechanism that allows wombats to produce cube-shaped poop.

AI/ML arXiv cs.AI

SwiftAudio: Data-Efficient Caption-Only Distillation for One-Step Text-to-Audio Diffusion-based Generation

SwiftAudio is a one-step text-to-audio diffusion framework that performs audio-free distillation using only text captions to reduce inference latency.

AI/ML arXiv cs.AI

TDGT: A Tabular Data Generation Toolkit supporting adaptive GPU-accelerated Bayesian mixture models, diffusion-based models, and latent-space generative modeling

TDGT is a web-based toolkit for synthetic tabular data generation and fidelity assessment, featuring adaptive Bayesian mixture models and GPU acceleration.

AI/ML arXiv cs.AI

Learning from Failure: Inference-Time Self-Improvement for Computer-Use Agents

Research introduces a failure-driven self-improvement loop for computer-use agents, allowing them to learn from failed trajectories using LLM-based diagnosis and code patches.

AI/ML arXiv cs.AI

CLIMB: Centroid-Based Hierarchical Memory for Online Continual Self-Supervised Learning

CLIMB introduces a hierarchical centroid-based memory bank to improve online continual self-supervised learning by limiting representation drift.