The Archive

Distributed JEPA self-supervised framework for energy time-series forecasting via masked latent prediction across heterogeneous assets.

CLARE addresses continual learning scalability via sparsity-driven fine-tuning to mitigate task interference and plasticity loss in sequential multi-task settings.

SKIP framework reduces LLM inference overhead by guiding step-wise preference learning for concise reasoning without sacrificing accuracy versus full chain-of-thought.

Human-centered semantic validation framework for network traffic classification models to detect spurious pattern exploitation beyond predictive metrics.

ORDER framework dynamically routes heterogeneous queries to optimal RAG configurations via task-conditioned indexing and retrieval adaptation.

ThinkFlow proposes latent probabilistic memory for lifelong conversational agents using predictive coding instead of static textual memory pipelines.

FlexEE early-exit framework reduces per-token latency and weight movement costs in offloading-based LLM inference via layer-wise exit supervision.

UNRESTSENT200K dataset enables crisis sentiment analysis for Bangla during July 2024 Bangladesh uprising with 200K annotated social media comments.

PaperDoctor agent framework provides evidence-grounded pre-submission feedback for scientific papers with hierarchical evaluation across writing, code, theory, and experiments.

Study reveals LLM student assessment biases from explicit and implicit demographic signals, showing discrimination risks in educational scoring.

Autoformalizing argumentative inferences addresses gap between natural language reasoning and formal verification by reconstructing implicit warrants and commitments.

Nameless tokenization defends open-weight LLMs against control-token forgery, fixing vulnerability in all 256 audited chat tokenizers.

Study identifies structural negative transfer in federated graph neural networks when client graphs differ in topology, beyond label/feature distribution variance.

Interpretable causal forests algorithm predicts individual treatment effects while explaining heterogeneity for medicine and marketing applications.

Optimal control method improves target-language generation in multilingual models with less hyperparameter tuning than activation steering baselines.

HUMAID-NER dataset provides 60K disaster tweets with entity-level annotations for NER and event classification in humanitarian response contexts.

Affect-Prototype framework enables open-vocabulary multimodal emotion recognition with incomplete modal inputs using uncertainty-weighted fusion.

Flow model for antenna phase recovery and offset correction in millimeter-wave testing; domain-specific engineering, not frontier AI.

Latent-space semantic communication framework for decentralized UAV swarms using evidence injection; niche robotics application.

γOPD: temporal credit assignment for on-policy LLM distillation bridging token-level and sequence-level supervision tradeoffs.

RepoAtlas: multimodal repository navigation for LLM coding agents using evolving code graph views for issue resolution.

One-step distillation for point cloud cranial implant generation; medical imaging application with limited AI frontier relevance.

Study of local vs. global confidence signals in autoregressive LMs; implications for reliability and model oversight.

Low-Rank Quantile Surfaces causal discovery method; theoretical statistics contribution, not AI-systems focused.

Transformer-based limit order book forecasting for counterfactual market impact; financial ML, not frontier AI.

Text optimization detects and verbalizes subliminal learning effects in distilled models; safety implications for data poisoning.

Hyperbolic geometry framework for genomic sequence representation using Lorentz encoding; domain-specific bioinformatics.

OpenAI launches ads in ChatGPT; Amazon integrates ads; Walmart accepts Apple Pay—commentary on commercialization and platform friction.

When Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, wanted to assess AI’s impact on the economy over the next few years, she faced a long list of business and technical uncertainties. So she started with what she calls a “remarkable fact” that is not in question: A handful of so-called…

Multi-agent neural networks share information to reduce model complexity while maintaining performance on classification tasks.