AI/ML arXiv cs.AI

Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift

Benchmarking the robustness of foundation models for mammography under domain shift, identifying Mammo-FM and MaMA as strong performers.

AI/ML arXiv cs.AI

A Hyperbolic Neural Closure for M1 Radiation Transfer

A proposed hyperbolic neural closure for M1 radiation transfer to ensure stable and accurate simulations in radiative transfer problems.

Tech Business/VC Hacker News

IBM Stock has worst day

IBM stock experienced a significant decline, marking one of its worst trading days.

Tech Business/VC TechCrunch

Lucid Motors denies report it’s considering bankruptcy

Lucid Motors has denied reports that it is considering bankruptcy following a sharp drop in its stock price.

AI/ML arXiv cs.AI

When Does Depth Survive Composition? Compute--Quality Regimes in Latent World Models

Researchers explore 'compute-quality regimes' in latent world models to determine when model depth actually improves prediction accuracy during autoregressive rollouts.

AI/ML arXiv cs.AI

Source-Lifted Flow Matching for Intervenable Multimodal Imitation

Introduction of Source-Lifted Flow Matching (SL-FM), a policy that allows users to intervene and choose between valid multimodal action continuations in imitation learning.

AI/ML arXiv cs.AI

Exploratory Analysis of Deep Learning Models for Forecasting Meteorological Parameters in the Agricultural Sector

A comparative study on using GRU, LSTM, and hybrid CNN architectures for forecasting meteorological parameters in agriculture.

AI/ML arXiv cs.AI

PhenoEmbed: Self-Supervised Multispectral UAV Time-Series Embeddings for Individual Tree Crown Phenology

PhenoEmbed is presented as a self-supervised model for creating temporal embeddings of tree crowns from UAV multispectral imagery to study phenology.

Software Engineering arXiv cs.AI

Partial Contracts Suffice: Sound, LLM-Inferred Regression Verification

A new contract-based regression verification tool that uses LLM-inferred partial contracts to ensure software patches preserve intended behavior.

AI/ML arXiv cs.AI

Program-Synthesis-Driven Autodesign of Universal Unitary Operators

Using program synthesis (extending DreamCoder) to automatically discover optimal decomposition strategies for unitary operators in photonic networks.

AI/ML arXiv cs.AI

Polarization Detection: A Hybrid Approach with AfroXLMR-Social and DeBERTa for Low- and High-Resource Settings

A hybrid approach using AfroXLMR-Social and DeBERTa for detecting polarized discourse in English and Hausa languages.

AI/ML arXiv cs.AI

Neutralizing Structural Inequality in the Nigerian FinTech Sector

A study on a hierarchical human-AI triage model to reduce structural inequality and bias in fraud detection for Nigerian FinTech.

Other Hacker News

The largest available Minecraft world, totalling 15 TB

Discussion regarding the largest available Minecraft world, which spans a total of 15 TB in size.

Tech Business/VC TechCrunch

Google faces another AI training lawsuit from major publishers

Google is facing new lawsuits from major publishers like Hachette and Elsevier for using copyrighted works to train its AI models.

AI/ML arXiv cs.AI

Transcript-Free Lightweight Detection of Alzheimer's Disease from Spontaneous Speech Using Handcrafted MFCC-Dominant Acoustic Biomarkers

Researchers developed a lightweight, audio-only SVM-based method to detect Alzheimer's disease from spontaneous speech without needing transcripts.

AI/ML arXiv cs.AI

Beyond Euclidean Clipping: Overcoming Exploration Collapse in LLM RL via Riemannian Isometric Policy Optimization

Introduction of Riemannian Isometric Policy Optimization (RIPO), which addresses exploration collapse in LLM reinforcement learning by using Riemannian geometry instead of Euclidean metrics.

AI/ML arXiv cs.AI

ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception

ActiveFly-Bench is a new benchmark for UAV embodied perception that bridges high-level reasoning with low-level aerial control.

AI/ML arXiv cs.AI

Automated Tensor Scheduling for Hybrid CPU-GPU LLM Inference on Consumer Devices

ATSInfer is a hybrid CPU-GPU inference system that improves LLM throughput on consumer devices by scheduling offloading at the tensor level.

AI/ML arXiv cs.AI

PhysMRV: Physical Memory Retrieval and Verification for Physics Plausibility Reasoning

PhysMRV is a training-free framework that uses structured physical memory banks to improve the physical reasoning capabilities of video-language models.

AI/ML arXiv cs.AI

Breaking the Quality--Intelligibility Trade-off in Streaming Target Speaker Extraction via Deep-Feature-Anchored Preference Optimization

A new method for Target Speaker Extraction uses WavLM-anchored Direct Preference Optimization to break the trade-off between audio quality and speech intelligibility.