AI Paper Tracker

Diversity boost in CV: LatinX in CV (LXCV) workshop at CVPR 2026 connects LatinX faculty, students, and engineers worldwide via speakers, orals, and...

TurboQuant breakthrough compresses LLMs and vector search 6x in working memory, enabling efficient inferencing without accuracy loss.

• Core tech:...

Robust ML eval guide from reproducibility study:

• XAI methods outperform non-XAI baselines on synthetic/real data
• CFKD excels for consistent...

Trending ML-adjacent quantum papers shake crypto foundations:

• Shor's algorithm for factorization/discrete logs viable with 10,000 reconfigurable...

Key trend in trustworthy, improvable AI agents:

• LangChain's three-layer framework (model, harness, context) shifts learning beyond weights, powering...

PRAM delivers parameter-free domain adaptation for frozen ICU models by retrieving outcomes from similar local patients.

• Simpler base models...

Key angles on ICML 2026 amid AI boom:

• Workshop explosion: 247 proposals (up from 150) for 44 slots, with agentic AI in 60 titles; chairs...

New arXiv paper introduces PLUME, a latent reasoning-based universal multimodal embedding designed for multimodal reasoning. Join the discussion on the paper page.

New automated system boosts interactive media environments:

• Combines ZED 2i depth camera, YOLOv10x tracking, and GPT-4o visual reasoning to detect...

Meta's EUPE family delivers vision encoders under 100M parameters that match specialists in image understanding, dense prediction, and VLM...

New paradigm shift for LLM training:

• Mainstream LLMs use Next-Token Prediction (NTP)
• Multi-Token Prediction (MTP) enables multiple tokens per...

Key findings from new blog on medical AI chatbot safety:

• Limited data exists on harms to real patients from AI chatbots.
• Incident reports, clinical trials, and health provider data show bots are usually safe.
• But they can be harmful in a few cases.

GAI reshapes DS curricula by automating routine workflows like cleaning, visualizing, and modeling.

• Human essentials endure: Problem formulation,...

New hybrid framework improves NLP baselines:

• Combines pre-trained BERT for contextual embeddings with BiLSTM to capture sequential dependencies in...

• New track for position papers at NeurIPS 2026, ideal for opinion pieces.
• Held to standard NeurIPS scholarship expectations.
• Requires evidence and reasoning to support claims, plus inclusion of key elements.

ONE-SHOT advances one-shot generation for dynamic human-environment videos using spatial-decoupled motion injection and hybrid context integration. Perfect for broad AI video synthesis tracking.

GR3EN brings generative relighting to 3D environments, granting full control over every light source—add new ones, tweak colors, or turn them off—while accurately handling shadows and reflections.

New paper reveals test-time scaling makes overtraining compute-optimal, optimizing inference efficiency in AI training.

New paper spotlights the Geometric Alignment Tax, pitting tokenization against continuous geometry in scientific foundation models. Key debate on discretization challenges—join the discussion.

Graph attention networks mine chemical abundance space by using graphs as the most suitable data representation, storing chemical abundances directly.