AI/ML arXiv cs.AI

RealDESED: A Real-World Domestic Sound Event Detection Benchmark

RealDESED is introduced as a real-world domestic sound event detection benchmark to bridge the gap between simulated datasets and real-world deployment.

AI/ML arXiv cs.AI

Spatiotemporal Facial Action Unit Detection using Twin Cycle Autoencoders for Driver Monitoring

The Twin Cycle Autoencoder (TCA) is proposed for spatiotemporal facial action unit detection to monitor driver drowsiness and distraction on embedded platforms.

AI/ML arXiv cs.AI

JOR-Bench: Japanese Operations Research Benchmarks for Large Language Models

JOR-Bench provides five Japanese-language benchmarks to evaluate LLMs' ability to formulate and solve operations research problems.

AI/ML arXiv cs.AI

Explainable Lightweight Compact Deep Models for Speech Emotion Recognition

A framework for explainable and lightweight speech emotion recognition is presented, utilizing a compact CNN and Grad-CAM for transparency.

AI/ML arXiv cs.AI

First-Order Predictable but Pairwise Fragile: Local Task Adaptation in Trained Transformers

Research on local task adaptation in Transformers reveals that while first-order perturbations are predictable, pairwise task composition is fragile and order-sensitive.

Other Hacker News

Recreating the math behind the first stealth aircraft

A discussion on recreating the mathematical foundations used to design the first stealth aircraft.

Other Ars Technica

Range Rover answers the question: "What if we built a not-SUV?"

Range Rover announces the debut of the Range Rover GT, an electric vehicle with a hybrid version to follow.

Other Ars Technica

On the run for 20 years, most-wanted fugitive caught hiding as a biotech exec

A most-wanted fugitive was captured in New York after hiding as a biotech executive for 20 years.

Other Ars Technica

Sony releases one last trailer for Spider-Man: Brand New Day

Sony releases a final trailer for the movie Spider-Man: Brand New Day.

Other Ars Technica

Confusion swirls on source of diarrhea outbreak, but it’s still Taylor Farms

Continued confusion regarding the source of a diarrhea outbreak linked to Taylor Farms.

AI/ML arXiv cs.AI

Privacy Cost as Equity Input: A Group Fairness Criterion for Differentially Private Machine Learning

Proposes the Privacy-Cost Equity Ratio (PCER) to ensure fairness in differentially private machine learning by accounting for privacy costs as equity inputs.

Hardware/Chips arXiv cs.AI

CLOSER-Bench: Evaluating Budgeted Cross-Stage Design Closure for Hardware Agents

Introduces CLOSER-Bench, a protocol for evaluating hardware agents' ability to achieve design closure across RTL-to-GDS stages.

AI/ML arXiv cs.AI

TellTale: Blending Multi-Instance LoRA Text Encoders and a Zero-Shot LLM Judge for Ambivalence/Hesitancy Recognition in Videos

Presents TellTale, a text-only approach for recognizing ambivalence and hesitancy in videos using LoRA-tuned text encoders and a zero-shot LLM judge.

AI/ML arXiv cs.AI

Falsification-Based Verification of LLM-Generated Optimization Models: Sound Test Batteries and Their Detection Limits

Develops a theory and test battery for the falsification-based verification of optimization models generated by LLMs.

AI/ML arXiv cs.AI

Position: Explanation Stability Is a Property of the Model Method Pair, Not the Model

Argues that explanation stability in AI is a property of the model-method pair rather than the model alone, calling for cross-method validation.

Hardware/Chips Hacker News

An Engineer's Guide to USB Type-C [pdf]

A comprehensive technical guide focusing on the specifications and engineering details of the USB Type-C standard.

AI/ML Hacker News

Gemini last models: temperature, top_p, and top_k are deprecated and ignored

Google's Gemini last models have deprecated temperature, top_p, and top_k parameters, rendering them ignored in API calls.

Cybersecurity Hacker News

Furtex: Post-exploitation, rootkit and evasion research toolkit for Linux

Furtex is introduced as a research toolkit for Linux post-exploitation, rootkit development, and evasion techniques.

AI/ML arXiv cs.AI

Building2Building: A Large Scale Benchmark for Generalizable Real-World Reinforcement Learning

Introduction of Building2Building (B2B), a large-scale benchmark for HVAC control to study generalizable reinforcement learning in real-world settings.

AI/ML arXiv cs.AI

Capacity and Redundancy Trade-offs in Multi-Task Learning

Research on Capacity and Redundancy Trade-offs in Multi-Task Learning, proposing a Capacity-Redundancy identity to analyze negative transfer.