PathFM Uses Flow Matching for Trajectory Forecasting, Tops TimesFM-3
autogluon · hf · 2026-10-17
The AutoGluon team releases PathFM, a multivariate time-series forecasting model that uses flow matching to generate complete future trajectories instead of step-wise point forecasts, enabling efficient multi-path sampling and better dependence modeling.
PathFM-300M scores 0.4735 on FEV and 0.5458 on FPS, beating TimesFM-3 and Chronos-2. On 4,096-step Lorenz attractor continuation, Chronos-2 and PANDA collapse to a flat mean while PathFM preserves the attractor structure.
More from Research
- SYNTH team's jspace analysis: Baguettotron layer topology radically differs from Gemma-270m — voooooogel · 2026-10-17
- Tangermeme published in Nature Methods: an 'everything-but-the-model' toolkit for genomic deep learning — anshulkundaje · 2026-10-17
- NYU's Qi Lei and Pavel Izmailov hiring postdoc on self-improving AI and AI for math — canondetortugas · 2026-10-17
- Context Inertia: when LLM agents latch onto a wrong hypothesis and won't let go — rseroter · 2026-10-17
- Microsoft's MAI-Base-2 blog source code leaks intermediate checkpoint validation losses — YouJiacheng · 2026-10-17
- Axolotl v0.21.0 ships 501B MoE QLoRA fine-tuning on a single 8xH200 node — ethayarajh · 2026-10-17