PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation
Summary
Flow-matching generators often smooth away sharp temporal patterns. PrismFlow splits the dynamics into a base flow plus expert residuals that specialize on different regimes. If you synthesize time-series data for testing or privacy, this offers more realistic signals without exploding model size.
Related Content
huggingface/transformers
The standard library for state-of-the-art models in text, vision, audio, and combined formats. If you build with open models, you almost certainly depend on this already.
HuggingFace's Transformers: State-of-the-art Natural Language Processing
This 2019 paper launched the Transformers library, giving a clean API around many transformer models and pretrained checkpoints. It turned cutting-edge NLP into a reusable software layer that underpins most open-source LLM work today.
SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning
The authors build SpatialClaw, a code-driven agent that uses a stateful Python kernel plus vision tools to solve 3D and 4D spatial puzzles. It beats prior spatial agents across 20 benchmarks and six vision-language backbones, showing that the action interface design can unlock much stronger spatial reasoning.
ENPIRE: Agentic Robot Policy Self-Improvement in the Real World
Wraps real robots in a closed-loop system where coding agents iteratively reset scenes, run policies, check results, and improve code. If you’re serious about autonomous robot labs, this is basically a blueprint.