Watch 9 videos detailing the real-world capabilities of the Gemini Omni and Gemini 3.5 models announced at Google I/O 2026.
Sponsored
Visit
Trending
Popular
Models
Selective Sinkhorn Routing: Efficiently Optimizing Sparse Mixture-of-Experts Routing
This paper introduces Selective Sinkhorn Routing (SSR), a lightweight method for assigning tokens in Sparse Mixture-of-Experts (SMoE) models. SSR optimizes expert utilization through optimal transport constraints, significantly improving model performance and training efficiency.
Research
Theoretical Link Between Scaling Laws and Weight Spectra in Shallow Neural Networks
This research establishes a theoretical framework for understanding scaling laws in neural networks by linking them to the spectral properties of trained weights, providing a first-principles interpretation for empirical observations regarding generalization performance.
Tools
Security Validation Now Available for Third-Party Coding Agents on GitHub
GitHub has introduced general availability for security validation for third-party coding agents, allowing developers to safely utilize tools like Claude and OpenAI Codex within their repositories.
Research
Robust Causal Discovery in Time Series via Power-Law Spectral Features
A new method for Causal Discovery (CD) in noisy real-world time series by leveraging the inherent power-law distribution observed in frequency spectra, leading to more robust causal inferences.
Research
Conformal Risk Sharing: Certified Cost Allocation for Sharing Adverse Event Impacts
A new framework, Conformal Risk Sharing, provides a mathematically certified method for allocating financial burdens resulting from rare adverse events across multiple participants, ensuring that no individual is made materially worse off while bounding aggregate harm.
Models
Boosting Brain-to-Image Decoding with Large Model Data Augmentation
This research investigates how augmenting small fMRI datasets with synthetic data from a large, pre-trained model (TRIBE v2) can significantly improve the accuracy of image decoding, showing up to a 68% gain over real-data-only training.