Salesforce AI Open-Sources ‘OmniXAI’: A Python-based Machine Learning Library That Provides One-Stop Explainable AI (XAI) Solution To analyze, Debug, And Interprets AI Models

Artificially Intelligent models, particularly those based on deep neural networks, have been viewed as black-box models since the rise of machine learning. This is mainly because of their complexity and lack of human understandability. As a result, some clinicians are hesitant to deploy AI models in certain crucial applications, even though their use could be advantageous. Because of these constraints, the topic of Explainable AI, also known as XAI, has seen a surge in attention. XAI aims to investigate and reveal the reasons behind an AI model’s decisions. It attempts to see what is inside the black box by opening it. These explanations can be the first step in increasing AI systems’ transparency and persuasiveness, aiding AI developers in debugging and improving model performance. Several popular XAI techniques have arisen, each with its own set of distinctive talents and characteristics.

On the other hand, most existing XAI libraries only handle a restricted set of data types and models. Furthermore, because different libraries have highly varied interfaces, switching from one XAI library to another while looking for a new explanation method becomes inconvenient for researchers. There is also a need for a visualization tool for users to examine and compare explanations, which is lacking in most existing solutions. To address these flaws, Salesforce has built an open-source machine learning framework called OmniXAI, which stands for Omni eXplainable AI. This library takes an “omni-directional” approach to XAI, with extensive interpretable ML features that address many problems with explaining ML model decisions in reality. OmniXAI is a one-stop comprehensive library that makes explainable AI accessible to academics requiring explanations for each stage of the machine learning process. This is not limited to data exploration, feature engineering, model development, evaluation, decision making, etc. 

OmniXAI offers a wide range of explanation methods divided into “model-agnostic” and “model-specific” categories. “Model-agnostic” refers to a method that can explain a black-box model’s decisions without understanding the model’s details, whereas “model-specific” refers to a method that requires some knowledge of the model to generate explanations. OmniXAI also offers feature analysis and selection, including looking at feature correlations, checking for data imbalances, and choosing features based on mutual information. It works with the most popular machine learning frameworks and models, including PyTorch, TensorFlow, and others. OmniXAI simplifies a researcher’s job by allowing them to use it for tabular, vision, NLP, and time-series tasks. In practice, users select from various explanation methods to examine various elements of AI models. OmniXAI makes this process easier for users by offering a consistent interface that allows them to generate multiple types of explanations with only a few lines of code. OmniXAI’s central design philosophy is to allow users to simultaneously apply several explanation methods and visualize the resulting generated explanations. New explanation methods can be quickly added without altering the library foundation, making the library very easy to use and develop.

Source: https://arxiv.org/pdf/2206.01612.pdf

The developers believe that putting an XAI library like OmniXAI to use in the actual world will benefit AI and humanity. The lack of explainability in AI models hampered confidence and stifled adoption in fields such as healthcare and finance. The judgments made by AI models may now be explained thanks to libraries like OmniXAI. This will improve AI systems’ transparency and persuasiveness, allowing customers to comprehend the reasoning behind their judgments. OmniXAI can also be helpful in situations where an AI model fails because it can provide answers to why the model failed. This will aid developers in swiftly determining the causes of failures and improving their models. OmniXAI is being continuously developed and improved by the Salesforce team, with more algorithms for feature analysis and support for various data types and tasks added. 

This Article is written as a summary article by Marktechpost Staff based on the paper 'OmniXAI: A Library for Explainable AI'. All Credit For This Research Goes To Researchers on This Project. Checkout the paper, article, github, dashboard.

Please Don't Forget To Join Our ML Subreddit

Khushboo Gupta is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Goa. She is passionate about the fields of Machine Learning, Natural Language Processing and Web Development. She enjoys learning more about the technical field by participating in several challenges.