TruEra, provider of a suite of quality AI solutions, launches TruLens, an open source explainability software tool for machine learning models based on neural networks.
TruLens is a library for deep neural networks that provides a uniform API for explaining Tensorflow, Pytorch and Keras models. The software is available for free download and comes with documentation and a community of developers for further development and use.
The library provides a cohesive and cohesive approach to explaining deep neural networks based on published research.
It natively supports internal explanations that bring up important concepts learned by units in the network, for example showing which visual concepts in images a facial recognition model uses to identify people or a radiological diagnostic model uses to identify medical conditions.
The library is based on a series of published academic articles. A set of key ideas emerge from the article Influence-Driven Explanations for Deep Convolutional Networks written by the creators of the Carnegie Mellon University Library.
The library also supports a set of other popular explainability techniques created by the research community, including saliency maps, built-in gradients, and SmoothGrad, which are widely used in computer vision use cases. and natural language processing.
TruLens has been used in a wide variety of real-world use cases to explain deep learning models. Use cases for neural network models include:
- Computer vision: identify a person, animal or object in a series of photos; categorize types of damage for insurance claims or review medical images
- Natural language processing: identification of malicious speech, analysis of social media posts, predictive text or smart assistants
- Forecasting: Using multiple inputs, including text and number inputs, to predict future events, such as probabilities of financial results
- Personalized recommendations: using past behavior to predict a user’s interest in other products
TruLens provides the ability to accurately explain the performance of these models, which allows developers to better understand and refine their models during the development phase, as well as to assess the performance of current models and correct models. once they are used in the real world.
“TruLens reflects the eight years of explainability research this team has developed at both Carnegie Mellon University and TruEra,” said Anupam Datta, Co-Founder, President and Chief Scientist of TruEra. “This means it starts off as a robust, focused solution with a strong lineage. There is also a team of very knowledgeable people ready to help developers explore how to use TruLens. We look forward to building an active developer community around TruLens.
For more information on this news, visit https://truera.com.