The Archive

Residual-space adaptation enables continual personalization in diffusion models, learning sequential concepts without catastrophic forgetting.

Analysis of cluster collapse in federated learning and control-theoretic interventions for stable client routing.

Multi-horizon time-series forecasting study using LightGBM on Google Trends and COVID-19 data with entropy-based error metrics.

OptScientist framework discovers symbolic optimization laws from data using neural inference and LLM-agentic exploration.

Theoretical convergence analysis of SGD in deep linear neural networks with regularized squared loss.

AnomalyAgent ensemble framework uses agentic fusion of vision-language models for anomaly detection with adaptive weighting.

Lafa: 8B-parameter flow language model with blockwise causal-bidirectional attention enabling parallel generation.

Empirical study comparing LLM-based and classical feature selection on tabular data with GRPO-adapted Qwen and Llama models.

CSFRBench measures gradient-based learning of compositionally sparse functions to reduce sample complexity via structure discovery.

Meta's AI companion Muse has surpassed 10 million U.S. downloads in its first 40 days, joining an exclusive group of apps that includes Threads. New data also shows Muse is retaining users at higher rates than its AI rivals.

Freshness-weighted gossip aggregation protocol for decentralized federated learning handling client outages and stale local models.

Binary Discrete Diffusion enables partial-uncertainty representation in discrete generative models for reasoning and constrained coding tasks.

Debate-based AI oversight fails when weaker judges face deceptive arguments from capable models, limiting the Safety via Debate approach.

Vector quantization and calibration data selection enable 2-bit post-training quantization of large models while maintaining usability.

REACT framework uses VLM-guided contact modeling to learn humanoid-human interaction policies from motion-capture data despite occlusion.

DecisionPivot benchmark evaluates routing, guardrail, and verification models in LLM systems across 48 tasks with 40K test items.

GraphCast-Mamba adds learnable Mamba correction module to frozen GraphCast backbone to capture multi-step temporal dependencies in weather forecasting.

Theoretical analysis of variance-reduced stochastic optimization under delayed gradient feedback with oracle complexity bounds.

Multimodal deep learning predicts laboratory values (NT-proBNP, glucose) from ECG recordings and patient metadata.

BitMTP combines ultra-low-bit quantization with multi-token prediction for efficient LLM decoding without quality loss.

Minimal-pair probing reveals latent information exists in transformer models but remains unreadable to standard single-prompt probes.

Closed-loop forecasting with spiking neural networks and state-space models for time-series prediction with temporal dependencies.

OptionWorld benchmark reveals LLMs fail at multi-day task completion, with 35.5% of failures due to lost context before deadline.

AsymLoop separates context encoding from iterative refinement in looped Transformers to improve effective depth without scaling parameters.

Physics-informed neural framework discovers extreme scenarios via data-driven dynamics inference with geometry-agnostic learned representations.

Proof-Carrying Contracts framework enforces agent compliance with verifiable commitments using formal constraints, enabling trustless multi-agent economies.

G-JEPA enhances latent world models by incorporating temporal context into physical grounding and improving action planning in JEPA frameworks.

Information decomposition provides formal framework for operationalizing emergence and synergy claims in AI systems beyond informal assessments.

User-location graph approach generates synthetic human mobility trajectories while preserving evolving community structures in datasets.

It’s the moment everyone is waiting for — the ChatGPT moment for robotics, Amulya Vishwanath explained to TechCrunch’s Equity.