Other Hacker News

Scientists find molecular-level evidence for two structures in liquid water

Scientists have discovered molecular-level evidence suggesting that liquid water contains two distinct structures.

Software Engineering Hacker News

Kb – Prolog Knowledge Base

Introduction of Kb, a knowledge base system built using Prolog.

AI/ML arXiv cs.AI

Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs

GRACE is a framework that combines knowledge distillation and quantization-aware training to create efficient INT4 Vision-Language Models with minimal accuracy loss.

AI/ML arXiv cs.AI

Spectral Text Fusion: A Frequency-Aware Approach to Multimodal Time-Series Forecasting

SpecTF is a frequency-aware framework that integrates textual embeddings into the frequency domain for improved multimodal time-series forecasting.

Cybersecurity arXiv cs.AI

When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models

Researchers introduce VJA, a visual-to-visual jailbreak attack on image editing models, and propose a training-free defense mechanism based on multimodal reasoning.

AI/ML arXiv cs.AI

Can Generative Artificial Intelligence Survive Data Contamination? Theoretical Guarantees under Contaminated Recursive Training

A theoretical study providing guarantees that generative AI models can converge to true data distributions even under recursive training with contaminated data.

AI/ML arXiv cs.AI

An Interpretable, Controllable Time-Varying IIR Denoiser for On-Device Assistive Hearing

TVF is a low-latency, interpretable speech enhancement model utilizing a cascade of IIR filters for on-device assistive hearing.

AI/ML arXiv cs.AI

MPFlow: Multi-modal Posterior-Guided Flow Matching for Zero-Shot MRI Reconstruction

MPFlow is a zero-shot multi-modal MRI reconstruction framework that uses rectified flow and cross-modal guidance to reduce hallucinations.

AI/ML arXiv cs.AI

Measuring the Redundancy of Decoder Layers in SpeechLLMs

Research on SpeechLLMs shows that decoder layers are highly redundant, allowing for significant pruning without substantial loss in ASR performance.

AI/ML arXiv cs.AI

EXPLORE-Bench: Egocentric Scene Prediction with Long-Horizon Reasoning

EXPLORE-Bench is introduced as a benchmark for measuring long-horizon egocentric reasoning in multimodal large language models for embodied agents.

Open Source Hacker News

Free the Icons

A discussion regarding the liberation or open-sourcing of icons.

Other Hacker News

Is It Out Yet?

A thread discussing the anticipation and release status of an unspecified product or feature.

AI/ML arXiv cs.AI

Hybrid Fact-Checking that Integrates Knowledge Graphs, Large Language Models, and Search-Based Retrieval Agents Improves Interpretable Claim Verification

Introduces a hybrid fact-checking pipeline integrating Knowledge Graphs, LLMs, and search agents to improve claim verification interpretability and accuracy.

AI/ML arXiv cs.AI

Hybrid coupling with operator inference and the overlapping Schwarz alternating method

Proposes a hybrid coupling method using operator inference and the overlapping Schwarz alternating method to speed up 3D solid dynamics simulations by up to 106x.

AI/ML arXiv cs.AI

Trust Region Masking for Long-Horizon LLM Reinforcement Learning

Presents Trust Region Masking (TRM) to provide non-vacuous monotonic improvement guarantees for long-horizon LLM Reinforcement Learning.

AI/ML arXiv cs.AI

Pixelwise Uncertainty Quantification of Accelerated MRI Reconstruction

A framework for pixel-wise uncertainty quantification in accelerated MRI reconstruction using conformal quantile regression to identify unreliable image regions.

AI/ML arXiv cs.AI

Psychometric Comparability of LLM-Based Digital Twins

Evaluates the psychometric comparability of LLM-based digital twins against human standards, finding they are most effective within specific validated boundaries.

Cybersecurity arXiv cs.AI

DDSA: Dual-Domain Strategic Attack for Spatial-Temporal Efficiency in Adversarial Robustness Testing

Introduces DDSA, a resource-efficient adversarial robustness testing framework that optimizes testing through temporal selectivity and spatial precision.

AI/ML arXiv cs.AI

Reasoning-Enhanced Rare-Event Prediction with Balanced Outcome Correction

Proposes LPCORP, a two-stage framework combining reasoning-enhanced prediction and confidence-based correction to improve rare-event prediction in imbalanced datasets.

AI/ML arXiv cs.AI

Robustness of Constraint Automata for Description Logics with Concrete Domains

An automata-based approach to prove the EXPTIME upper bound for the consistency problem of description logics with concrete domains.