Cybersecurity arXiv cs.AI

Mitigating The Effect of Class Imbalance in Data with Hierarchical and Dependable Structure

Proposes a Hierarchy-Aware RoBERTa framework to improve CWE vulnerability classification by incorporating structural information.

AI/ML arXiv cs.AI

Are we Merging the Right Models? Impact of Expert Training Duration on Model Merging for LLMs

Studies how training duration of domain experts affects the quality of multi-task model merging for LLMs.

AI/ML arXiv cs.AI

HPC-Enabled Video-based Coastal Wave Parameter Estimation Using V-JEPA and Deep Spatiotemporal Learning

An HPC-enabled deep learning framework for estimating coastal wave parameters from monocular coastal video.

Other Hacker News

Who's running all those tiny RPKI servers?

A community discussion on Hacker News regarding the operational infrastructure and entities responsible for maintaining the numerous small Resource Public Key Infrastructure (RPKI) servers.

AI/ML arXiv cs.AI

BattVAE-GP: Generative Modeling of Long-Horizon Battery Degradation with Uncertainty Quantification

Introduction of BattVAE-GP, a hybrid physics-probabilistic learning framework for efficient, uncertainty-aware surrogate modeling of lithium-ion battery degradation trajectories.

AI/ML arXiv cs.AI

Generalized Distribution-Free Semi-Supervised Learning with Risk Rewrite

A generalized distribution-free semi-supervised learning framework using risk rewriting to provide unbiased risk estimators for multiclass classification with reduced variance.

AI/ML arXiv cs.AI

BAT-RM: A Boundary-Aware Transformer with Region-Aware Multi-Directional Mamba for Clinically Deployed Cervical Cancer Radiotherapy Auto-Contouring

BAT-RM, a boundary-aware Transformer and Mamba hybrid architecture, is deployed for cervical cancer radiotherapy auto-contouring, significantly reducing contouring time and improving consistency.

AI/ML arXiv cs.AI

Scale-Aware Attention for Scarce Neural Data: An RG-Flow Transformer on Sleep-EDF EEG

Evaluation of an RG-Flow Transformer with renormalization-group inductive bias for EEG sleep staging, showing limited performance gains but improved interpretability of spectral exponents.

AI/ML arXiv cs.AI

Graph-Constrained Policy Learning for Extreme Clinical Code Prediction

A graph-constrained traversal policy using language models to predict clinical ICD-10-CM codes, outperforming flat multi-label classification in extreme label spaces.

AI/ML arXiv cs.AI

Exact and Certified Data Shapley for Weighted k-Nearest-Neighbor Regression and Soft-Label Prediction

Introduction of the first pseudo-polynomial-time exact algorithm for weighted KNN-regression Data Shapley, including a certified FPTAS and an open-source CPU-only library.

AI/ML arXiv cs.AI

Evaluating Reliability in Machine Learning Models for Early Chronic Kidney Disease Prediction: A Systematic Review of Data Leakage and Predictor Stability

A systematic review of kidney disease prediction ML models, revealing that high data leakage often leads to inflated performance metrics and low predictor stability.

AI/ML arXiv cs.AI

Beyond Coordinate Gauge: An Audited Protocol for Detecting Donor-Specific Functional Fingerprints after Neural Collapse

An audited protocol for detecting donor-specific functional fingerprints in neural networks after neural collapse, establishing detectability under specific alignment mappings.

AI/ML arXiv cs.AI

Self-Evolving In-Context Learning for Direct Pilot-to-Beamformer Design in MU-MISO Systems

An enhanced in-context learning framework using a Transformer backbone for pilot-based beamforming in MU-MISO communication systems, improving adaptation to unseen channel models.

Other Hacker News

Surprising lessons from my research scientist job search

A personal account and discussion regarding the challenges and lessons learned during a research scientist job search.

Hardware/Chips The Verge

Starlink’s V5 dish is now available — here’s how it compares

SpaceX releases the Starlink V5 residential dish, featuring a smaller, lighter design and improved power efficiency over the V4.

AI/ML arXiv cs.AI

Burst Spiking Neural Networks

Introduction of Burst Spiking Neural Networks (BuSNNs), which improve accuracy and robustness in low-power SNNs through burst firing and dynamic weight constraints.

AI/ML arXiv cs.AI

QDEvo: A Multi-Objective Quality-Diversity Framework for Automated Heuristic Design

QDEvo is a new multi-objective framework that uses LLMs and quality-diversity evolution to design semantically diverse and high-performing heuristics for combinatorial optimization.

AI/ML arXiv cs.AI

AAAI-26 Dual Submissions: Novel Challenges

AAAI-26 organizers report a significant rise in dual submissions, often masked by generative AI, and propose new detection tools and policies.

AI/ML arXiv cs.AI

Do You Remember? Toward Memory-Centric Multimodal AI

The DoYouRemember architecture introduces reconstructive memory to MLLMs using VQ-VAE and a Diffusion Decoder to address the lack of visual memory in LLMs.

AI/ML arXiv cs.AI

Sensitivity to Subjective Expected Utility Maximization: A Methodological Study, with an Illustrative Application to LLM Decision-Making

A methodological study on subjective expected utility (SEU) maximization in LLM decision-making, analyzing how models like GPT-4o and Claude 3.5 Sonnet handle uncertainty.