Introducing Capsa

We are delighted to announce the launch of our Capsa Private Beta. On July 1st, we started our Private Beta, offered exclusively to 5 companies carefully selected from a waitlist of over 30. These pioneering partners represent some of the most innovative, visionary, and groundbreaking companies in the market. The choice of these partners was driven by their substantial machine learning expertise, ongoing active projects, and involvement in cutting-edge applications. With them we share mutual trust and a common vision for shaping the future of trustworthy AI together. 

We are also excited to announce that other companies will be able to join our waitlist as we are looking to add more companies to the Private Beta in the Fall. Though Capsa is for now only accessible by participating in the Private Beta, a wider release will be announced in the future.

Capsa is a Python library designed to infuse machine learning models with the ability to detect different levels of uncertainty. This product promises to revolutionize the way companies assess the safety and robustness of their models, thus advancing the goal of developing and adopting trustworthy artificial intelligence.  

Our library encompasses a range of methods cleverly packaged as easy-to-use "wrappers," making them applicable and scalable across any model. At any stage of model development (e.g., pre-training, post-training, during training, during inference), Capsa automatically converts models into an uncertainty-aware version, providing uncertainty estimates alongside every output. Capsa helps assess uncertainty along three main categories: aleatoric uncertainty (due to randomness and irreducible noise), epistemic uncertainty (due to lack of knowledge), and vacuitic uncertainty (due to gaps in the data).

Capsa’s main functionalities:

• Seamlessly integrate four classes of easy-to-use wrappers into any machine learning models to automatically evaluate uncertainty, and reliably predict failures within seconds.  
• Have access to an extensive library of cutting-edge uncertainty estimation types, algorithms, and techniques, tested on a number of very popular models.
• Adopt model-agnostic wrappers compatible with both Tensorflow and Pytorch models.

Founded in 2021 as a spin-off of DRL at CSAIL MIT, Themis AI is dedicated to harnessing evidence-based research and cutting-edge technology to empower the global community in creating, advancing, and deploying trustworthy AI.

Our journey began with extensive research on safe decision-making for autonomous agents. After years of development, we resolved to introduce our groundbreaking technology to the private sector with the goal of fostering trustworthy AI, especially in high-risk scenarios, rare events, and areas where unfair biases may lurk within the training data set. The implementation of Capsa is the result of over 7 years of MIT research, incorporating both published and proprietary algorithms developed by Themis AI.

Capsa’s capabilities have already made a significant impact, with successful applications in various fields such as autonomous driving, face detection, drug discovery, object detection, depth estimation, text encoding, and image generation. These accomplishments have garnered attention from the press and reinforced our commitment to drive innovation in AI technology.