Tag
Introduces CardioState-JEPA, a cardiac foundation model that learns a shared representation across ECG, PPG, and PCG signals using a delay-aware joint-embedding predictive architecture, improving downstream cardiac classification tasks.
This paper proposes a market-information-aware gated LoRA framework to adapt the Chronos-2 time-series foundation model for day-ahead electricity price forecasting, improving cross-market transferability on Chinese provincial markets.
Intern-S2-Preview is a scientific agentic foundation model series integrating multimodal pretraining, multi-task reinforcement learning, and memory-augmented extensions for long-horizon scientific reasoning and forecasting. The 397B model achieves competitive results across scientific and agentic benchmarks, with a separate memory-decoder extension improving biology instruction performance without modifying the backbone.
DoGMA is a central-dogma-guided foundation model for pan-cancer multi-omics analysis, using a Transformer-MoE architecture with directed attention to align DNA-RNA-protein flows and pretraining via masked hierarchical omics reconstruction. It shows strong performance across cancer representation learning, survival prediction, and metastasis prediction tasks.
Prodigy Research, a YC S26-backed startup, announces its launch as a frontier AI trading research lab, claiming to train a top-tier quantitative finance foundation model that outperforms top traders and major indices.
A user reacts to Meta's announcement of open-weights Muse Glimmer and upcoming Muse Spark 1.2, saying he'll apologize if Spark 1.2 is Apache 2.0 and genuinely great.
Meta is preparing to release the weights for Muse Spark 1.2, its latest foundation model, which will be available for broad use.
RynnValue introduces an open-source robotic value foundation model that uses temporal distance as a scalable supervision target for reward learning, surpassing preference-supervised state-of-the-art on RBM-EVAL-OOD and improving real-world policy success rates.
Researchers created VirTues, a unified foundation model for spatial proteomics that translates diverse tissue imaging data into a standard language, enabling faster and more accurate medical diagnoses and personalized treatments.
A new 1.1B-parameter DNA foundation model, MarinDNA v0.5 scaling ladder, was released on Hugging Face; it reads and generates DNA sequences and reportedly rivals Evo 2 40B on variant effect prediction.
This paper systematically evaluates 15 machine learning models, including the TabPFN foundation model, for post-wildfire debris-flow prediction using USGS basin-scale data, finding TabPFN achieves the best performance (threat score 0.637) and that synthetic data augmentation improves most models.
LG AI Research presents K-EXAONE 2.0, a 750B-parameter MoE foundation model upcycled from K-EXAONE, supporting 256K context and six languages, with self-speculative decoding for efficient inference.
This paper introduces TS2TabPFN, a framework that combines explicit feature extraction with the TabPFN 2.5 tabular foundation model for time series classification and extrinsic regression. Experiments show it outperforms state-of-the-art models in TSER and achieves competitive results in TSC.
Xiaomi open-sourced Xiaomi-Robotics-1, an embodied AI foundation model pretrained on over 100,000 hours of UMI data and post-trained on 10,000+ hours of cross-embodiment data. The release includes the full real-robot post-training and deployment pipeline, aiming to challenge proprietary robotics models from Figure AI and Tesla.
Xiaomi released XR-1, a robot foundation model trained on over 100K hours of real-world manipulation trajectories. Built on Qwen3-VL and a Diffusion Transformer, it enables out-of-the-box mobile manipulation in unseen environments.
K-EXAONE 2.0 is an open-weight multilingual MoE foundation model from LG AI Research with 750B total parameters and 37B active, supporting 10 languages and 256K context, with notable gains in agentic coding, long-context understanding, and safety.
Microsoft Research and Paige introduce PRISM2, a multimodal foundation model trained on pathology images and language, which matches specialized cancer-detection systems across benchmarks without task-specific models. The model weights are publicly available on Hugging Face for research.
This paper proposes DualIFM, an interpretable-by-design foundation model for retinal fundus images, achieving performance comparable to RETFound with far fewer parameters while providing interpretable predictions.
Memtensor Research Group released Metis, a family of LLMs (4B/9B/27B) that internalize memory into the backbone, eliminating external RAG. The model performs memory read/write in a single forward pass and deploys with frozen weights like a standard LLM.
Introduces ZUNA1.1, a 380M-parameter diffusion autoencoder for flexible EEG signal reconstruction, capable of handling variable-length sequences, arbitrary channels, and temporal intervals, while outperforming standard interpolation methods. The model is released open source under the Apache 2.0 license.