The Archive

Non-contrastive representation learning for detecting semantically equivalent code clones; addresses Type-IV clone detection challenges.

Self-Emergence Agent Architecture integrating HMM behavioral inertia, reflexive metacognition, and self-other boundaries for LLM agents.

Multi-judge ensemble approach for detecting hallucinated character spans in vision-language outputs; SHROOM-Visions shared task winner.

PosterVisor framework for scientific poster generation using persistent semantic-geometric contracts instead of transient prompts.

Tutorial on intrinsic motivation mechanisms (empowerment, curiosity, skill discovery) for adaptive self-organization in multi-agent neural systems.

ASURRE benchmark for detecting LLM-assisted survey responses; evaluates existing ML detectors against multiple generation strategies.

Profound has raised a $180 million Series D at a $1.8 billion valuation, less than seven months after it raised a $96 million Series C.

Zero-shot narrative detection in social messages using LLMs; human-written narrative descriptions improve classification without training examples.

Systematic evaluation of 8 model selection strategies for multi-agent systems; identifies gap between oracle and actual performance on scientific benchmarks.

Quantum-inspired tensor networks enable parameter-efficient image inpainting with O(N² log N) cost via diagonal QFT relaxation.

Goal-oriented probabilistic forecasting using Pinball Loss aligns 5G PRB allocation models with asymmetric operational costs.

Conformal policy learning framework provides distribution-free safety guarantees for high-stakes treatment allocation decisions.

LLM-based extraction of ontology-compliant knowledge graphs from scientific materials literature enables structured reuse.

Path choice in flow matching fundamentally affects SGD convergence rate despite identical FM objectives and marginal distributions.

Federated learning (FL) projects often begin with a straightforward setup: one server, a few clients, and one dataset at each site. As those projects grow, the... Federated learning (FL) projects often begin with a straightforward setup: one server, a few clients, and one dataset at each site. As those projects grow, the challenge shifts from running an algorithm to operating shared infrastructure. GPUs must be allocated when jobs need them, multiple research studies must remain separated, and every participating organization must retain control of its own… Source

Shortly after launching its new do-everything AI assistant Muse, Meta's launching subscription bundles that pair its standalone app subscriptions with extra AI usage. Some of the new Meta One bundles were in testing earlier this year, but are now available globally starting today, with several tiers for individual users, creators, and businesses. Meta says the "core experience" on its apps and Meta AI will still be free, and users can still get its subscriptions for Facebook, Instagram, and WhatsApp without a bundle. It also says it plans to expand the bundles to include "Edits, AI glasses, a...

pFedKDH improves personalized federated learning via global knowledge distillation and persistent client-specific heads under data heterogeneity.

OpenClaw viral agent-skill ecosystem case study reveals governance challenges as public LLM skill registries scale rapidly.

Essentially a camera app, Superpose analyzes selfies or photos and generates four potential poses using AI.

Reality gap in cyber-physical systems design stems from runtime environment divergence from design assumptions and simulations.

Persistent recurrent memory module between transformer layers reduces language model evaluation loss 28.5% with 3.7% parameter overhead.

Learn how to scale your startup breakthrough from prototype to production at TechCrunch Disrupt 2026 with scaling leaders, Adrian Macneil (Foxglove), John Mackey (MBRYONICS), and Boris Sofman (Bedrock Robotics. Register before September 25 to save up to $200 on your pass.

ECHO hierarchical dual-loop framework accelerates LLM inference via early-layer discrimination and stale draft mitigation in speculative decoding.

AraMIP framework extends MIPVU for Arabic metaphor annotation, addressing under-resourced linguistic datasets.

FROD applies image translation and denoising to assist oracle bone script decipherment via topological matching.

Study tests whether frozen LMs can diagnose corrupted reward signals using verified records, relevant to RL alignment.

Analysis of 100+ years of song lyrics reveals psychological distress signals during Vietnam War, 9/11, COVID-19.

Medical AI models exhibit memorization bias where predictions shift for patients whose historical data appeared in training.

SWE-Agents fail on Architecture-0 design phase with unknown unknowns; study shows text-only reasoning produces infeasible solutions.

FluxVLA Engine: open platform for VLA/WAM model training, eval, and deployment with unified data/runtime interfaces.