3D-Aware MTL w/ Cross-View Correlations revolutionizes dense scene understanding! Lightweight CvM injects geometric consistency via cost volumes for tasks like seg & depth—architecture-agnostic for single/multi-view data. Boosts perf on NYUv2/PASCAL! #ComputerVision #MTL #3D

3D-Aware Multi-Task Learning with Cross-View Correlations for Dense Scene Understanding https://t.co/v2fuvNNJ5T

POUR: Provably Optimal Unlearning of Representations via Neural Collapse. Forgets visual concepts at the rep level (not just classifier) w/ ETF projections, balancing forgetting, retention & separation. Outperforms SOTA on CIFAR/PathMNIST! #MachineUnlearning #NeuralCollapse

POUR: A Provably Optimal Method for Unlearning Representations via Neural Collapse https://t.co/0rNdsBRz0u

ShapeGen advances image-to-3D synthesis with improved 3D reps/supervision, higher res, & linear transformers—fixing smoothed surfaces & fragmented structures for artist-quality assets. Hits SOTA, pipeline-ready! #ShapeGen #3DGeneration #ComputerVision (187 chars)

ShapeGen: Towards High-Quality 3D Shape Synthesis https://t.co/0q776PfuoV

Genie-CAT: agentic LLM framework beyond protein LMs for mechanistic enzyme design. Integrates RAG, PDB parsing, electrostatics & ML redox predictions to generate testable hypotheses in metalloproteins—e.g., Fe-S cluster tweaks for redox tuning, faster than experts! #AI #Prot

Beyond Protein Language Models: An Agentic LLM Framework for Mechanistic Enzyme Design https://t.co/dWpv8klNVP

New open-source platform lets creators detect if their work is in LLM training data—user-friendly, scalable, & 10-30% more efficient than DE-COP. Enhances similarity checks & transparency for ethical AI dev. #AICopyright #GenerativeAI #EthicalAI #LLMs (187 chars)

Copyright Detection in Large Language Models: An Ethical Approach to Generative AI Development https://t.co/UDStENpC6y

New hybrid DL framework for lung cancer staging: segments anatomy (lobes, tumor, etc.), measures size/distances, applies rule-based TNM guidelines. Hits 91.36% accuracy on Lung-PET-CT-Dx (F1: 0.93 T1, 0.96 T3), beating black-box CNNs with interpretable decisions—first to emb

An Anatomy Aware Hybrid Deep Learning Framework for Lung Cancer Tumor Stage Classification https://t.co/jGb1tUnUw3

GUARDIAN: Gated Uncertainty-Aware Runtime Dual Invariants for safe neural signal-controlled robotics. Delivers 94-97% safety on EEG data w/ low-accuracy decoders (27-46%), 1.7x better interventions, 100Hz real-time monitoring. #BCI #Neuroscience #Robotics #AISafety (187 char

Gated Uncertainty-Aware Runtime Dual Invariants for Neural Signal-Controlled Robotics https://t.co/QuZmxE5ddP

MSTN revolutionizes multivariate time-series modeling with multi-scale conv encoder, seq modeling (BiLSTM/Transformer), & gated fusion to handle fast/slow dynamics & sudden events—beating fixed-scale priors. Achieves new SOTA on 24/32 benchmarks for forecasting, imputation &

MSTN: Fast and Efficient Multivariate Time Series Model https://t.co/sbGeLFD6tM

CAPNET unleashes CLIP for long-tailed multi-label recognition: models reliable label correlations via GCN & soft prompts, balanced Focal loss, test-time ensembling & PEFT to excel on tails w/o hurting heads. SOTA on VOC-LT, COCO-LT, NUS-WIDE! #ComputerVision #CLIP #MachineLe

Unleashing the Power of Vision-Language Models for Long-Tailed Multi-Label Visual Recognition https://t.co/daNA9DHTYa

Self-Paced GRPO revolutionizes video generation: rewards co-evolve with the generator, shifting from coarse fidelity to temporal coherence & semantic alignment. Fixes static reward biases for stable, superior training. Beats baselines on VBench! #VideoGen #AI #RLHF (187 char

Growing with the Generator: Self-paced GRPO for Video Generation https://t.co/TCaNDLQORb