AI/ML arXiv cs.AI

Multi-agent Autoformalization of Tensor Network Theory

Demonstrates an agent-driven workflow for the autoformalization of theoretical physics in Lean, specifically for tensor network theory, and releases the TNLean library.

AI/ML arXiv cs.AI

Time-to-Collision Based Dynamic Obstacle Avoidance Using Pretrained Vision Models for Robots in Unstructured Environments

Presents a data-efficient method for robotic dynamic obstacle avoidance using pretrained vision models (UniDepth) to compute time-to-collision without requiring extensive training.

AI/ML arXiv cs.AI

How Do I Know What to Say Next? Barenholtz's Autogenerative Theory as an Enrichment of Harrisean Integrationism

Discusses the synthesis of Integrationist linguistics and autogenerative theory to provide a structural account of how LLMs exploit statistical language patterns.

Software Engineering arXiv cs.AI

Closed-Loop Dynamic Validator Node Scaling in Private Substrate Blockchains Using Takagi-Sugeno Fuzzy Inference

Implements a Takagi-Sugeno fuzzy inference system for autonomous validator node scaling in private Substrate blockchains to optimize resource usage and performance.

Cybersecurity arXiv cs.AI

Mechanistic Interpretability of LLM Jailbreaks via Internal Attribution Graphs

Introduces a mechanistic interpretability framework using internal attribution graphs to diagnose and mitigate LLM jailbreaks by identifying vulnerability motifs.

AI/ML arXiv cs.AI

Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks

A comprehensive survey on multimodal unlearning across various AI models to selectively remove sensitive or unsafe cross-modal associations.

AI/ML Hacker News

AI-generated videos to maximally drive a target brain region

Research on using AI-generated videos to specifically target and stimulate brain regions for potential therapeutic or neurological effects.

Cybersecurity Hacker News

Browser Fingerprinting – How websites track you across internet –without cookies

An explanation of browser fingerprinting techniques used by websites to track users across the internet without relying on cookies.

AI/ML arXiv cs.AI

SHIFT: Survival Prediction from Incomplete and Heterogeneous Genomic Data

Introduction of SHIFT, a survival prediction model for genomic data that handles incomplete features without requiring test-time imputation.

AI/ML arXiv cs.AI

Collective Intelligence with Foundation Models

A study on a multi-agent framework for cooperative reasoning, finding that model heterogeneity is the primary driver for performance gains over homogeneous setups.

AI/ML arXiv cs.AI

Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE

Jet-Long is a tuning-free method for extending LLM context windows using dynamic bifocal RoPE, achieving high throughput and accuracy on long-context tasks.

AI/ML arXiv cs.AI

Architecture Generalization with MetaNCA

MetaNCA introduces a framework where neural cellular automata learn local rules to self-organize the weights of artificial neural networks without backpropagation.

AI/ML arXiv cs.AI

A Transdiagnostic Space of Disorder Like Phenotypes in Reinforcement Learning Agents

A framework for modelling psychological disorders in RL agents by manipulating cognitive appraisal signals to study affective phenotypes.

AI/ML arXiv cs.AI

Principled Analysis of Deep Reinforcement Learning Evaluation and Design Paradigms

A theoretical analysis of scaling laws and evaluation paradigms in deep reinforcement learning, challenging some canonical conclusions in the field.

AI/ML arXiv cs.AI

Graph-Regularized Deep Learning for EEG-Based Emotion Recognition with Psychologically-Grounded Label Structure

A graph-regularized deep learning framework for EEG-based emotion recognition that incorporates psychologically-grounded label structures.

AI/ML arXiv cs.AI

From Solvers to Research: Large Language Model-Driven Formal Mathematics at the Research Frontier

A position paper outlining the roadmap for transitioning LLM-driven formal mathematics from simple solvers to autonomous research agents.

Other Hacker News

Damaged Earth Catalog

A discussion thread on Hacker News regarding the Damaged Earth Catalog.

AI/ML arXiv cs.AI

The Illusion of Equivalency: Statistical Characterization of Quantization Effects in LLMs

Research demonstrating that traditional accuracy metrics fail to capture behavioral changes in quantized LLMs, proposing a new 'correctness agreement' metric.

AI/ML arXiv cs.AI

Workflow as Knowledge: Semantic Persistence for LLM-Mediated Workflows

A conceptual model for LLM-mediated workflows treating workflow definitions and instances as persistent, inspectable knowledge objects.

AI/ML arXiv cs.AI

AUTOPILOT VQA: Benchmarking Vision-Language Models for Incident-Centric Dashcam Understanding

Introduction of AUTOPILOT-VQA, a benchmark for evaluating the ability of vision-language models to reason about safety-critical dashcam incidents.