The Archive

Transfer learning method for physics-informed neural networks in PDE inverse problems; domain-specific technical advance.

RL-based adaptive batching and routing for inference serving outperforms static policies on production traces.

Audit of hyperbolic geometry in vision-language models identifies operating-point failure modes; questions practical utility.

Continuous reference biomass data from QSM improves LiDAR-based forest AGB estimation; specialized remote sensing application.

SalAngaBhava: aspect-based sentiment analysis dataset for Sinhala (low-resource language); dataset contribution.

GeoFlow augments graph neural networks with geospatial attributes for origin-destination flow prediction in urban mobility.

Government of Alberta deploys Claude Opus and Sonnet via Claude Code for cybersecurity vulnerability detection and remediation across government systems since 2025.

I’m just doing my job, but turns out, it feels lousy holding other people’s privacy in your hands. I've long argued that Hollywood has simultaneously set and ruined our expectations for smart glasses. But after binge-watching two seasons of Netflix's A Man on the Inside, this…

FUSE applies flow matching with Feynman-Kac steering for multimodal simulation-based posterior inference.

SecureCROWN enables privacy-preserving robustness verification for neural networks using secure two-party computation.

TiCodec separates time-varying and time-invariant speech content for low-latency neural codec applications.

CanniUplift addresses seller and incentive cannibalization in e-commerce uplift modeling via SUTVA violations.

ILP-based ML workload partitioning optimizes matrix-vector multiplication across CPU and RRAM-based CIM accelerators.

MoP-JEPA uses hard-assigned mixture-of-experts predictors to prevent regression collapse in stochastic world models.

Video-based machine learning detects apnoea-related breathing cessation in pre-term infants without contact sensors.

msPCA is an R package implementing sparse PCA with multiple non-redundant components for high-dimensional data.

Iterative pseudo-labeling improves Mandarin-English code-switching ASR by leveraging unlabeled data via semi-supervised training.

Expert evaluation of Evrópuvefur, Iceland's government LLM service for EU referendum Q&A, measuring curated vs. web-search source trustworthiness.

Probing self-supervised learning representations in satellite imagery against ERA5 environmental variables to assess information encoding.

In the AI era, platforms have no choice but to fight fire with fire to cull spam.

EvoAgentBench: benchmark for LLM agent self-evolution measuring procedural ability transfer across web research, reasoning, software engineering, knowledge work.

FlatManifold: continual learning framework using kernel manifold flattening to handle label noise and domain shifts in streaming environments.

Step-level reward framework with structured feedback trains small LMs to correct physics reasoning errors via policy gradient without ground-truth targets.

Noisy-channel approach to Minimum Bayes Risk decoding incorporates bidirectional metric asymmetry for improved text generation quality.

Nemotron-Labs Audex-30B: unified audio-text MoE LLM enabling seamless multimodal generation via single Transformer decoder with shared embedding space.

Stealth memory injection attack on persistent personal agents: adversary injects poisoned data via untrusted content, remains hidden, affects future behavior.

Microsoft cut around 4,800 roles, or 2.1% of its global workforce, on Monday — the latest in a series of layoffs that’s stoking fears of AI replacing jobs. The layoffs will hit Xbox and commercial sales the hardest.

Residual streams in coding agents linearly encode program properties (parse status, test results, regressions) up to 0.83 AUC.

SMART: ML+Monte Carlo framework for rapid transistor aging and process-variation analysis in deep-nanometer CMOS designs.