Timestral: Mistral release time-series foundation models

Timestral, a company founded by researchers from the University of Cambridge, has released a set of time-series foundation models that can be used for tasks such as anomaly detection and forecasting. The models, called Timestral, are pre-trained on a large dataset of time-series data and can be fine-tuned for specific use cases. Timestral's models are based on a combination of transformer and convolutional neural networks, and are designed to handle long-range dependencies in time-series data. The company claims that its models outperform other state-of-the-art models on several benchmark tasks.