Artificial intelligence (AI) has been integrated into every part of our lives. A chatbot, enabled by advanced Natural language processing (NLP), pops to assist you while you surf a webpage. A voice recognition system can authenticate you in order to unlock your account. A drone or driverless car can service operations or access areas that are humanly impossible. Machine-learning (ML) predictions are utilized to all kinds of decision making.

Interpretable representations are the backbone of many black-box explainers. They translate the low-level data representation necessary for good predictive performance into high-level human-intelligible concepts used to convey the explanation. Notably, the explanation type and its cognitive complexity are directly controlled by the interpretable representation, allowing to target a particular audience and use case. However, many explainers that rely on interpretable representations overlook their merit and fall back on default solutions, which may introduce implicit assumptions, thereby degrading the explanatory power of such techniques. To address this problem, we study properties of interpretable representations that encode presence and absence of human-comprehensible concepts. We show how they are operationalised for tabular, image and text data, discussing their strengths and weaknesses. Finally, we analyse their explanatory properties in the context of tabular data, where a linear model is used to quantify the importance of interpretable concepts.

Robots have moved off the assembly line and into warehouses, offices, hospitals, retail shops, and even our homes. ZDNet explores how the explosive growth in robotics is affecting specific industries, like healthcare and logistics, and the enterprise more broadly on issues like hiring and workplace safety. But machine learning (ML), which many people conflate with the broader discipline of artificial intelligence (AI), is not without its issues. ML works by feeding historical real world data to algorithms used to train models. ML models can then be fed new data and produce results of interest, based on the historical data used to train the model.

"My dog accidentally knocked down the trash and found old cheesy pasta in it, and is now convinced that trash cans provide an endless supply of cheesy pasta, knocking it over every chance she gets." Sometimes, you would have seen your Machine Learning(ML) model do the same too. A notorious example is how a neural network learned to differentiate between dogs and wolves. It didn't truly learn to differentiate between dogs and wolves, instead it learnt that all the wolf pictures had snow in the background as it was their natural habitat as opposed to dogs with grass in the background. The model then differentiated the two animals by looking at whether the background was snow or grass.

I will keep this section short but informative because as a Data Scientist we know very well why do we need to explain our models. Lets understand the need of XAI form below scenario.

Recently there has been an increasing interest in frameworks extending Dung's abstract Argumentation Framework (AF). Popular extensions include bipolar AFs and AFs with recursive attacks and necessary supports. Although the relationships between AF semantics and Partial Stable Models (PSMs) of logic programs has been deeply investigated, this is not the case for more general frameworks extending AF. In this paper we explore the relationships between AF-based frameworks and PSMs. We show that every AF-based framework $\Delta$ can be translated into a logic program $P_\Delta$ so that the extensions prescribed by different semantics of $\Delta$ coincide with subsets of the PSMs of $P_\Delta$. We provide a logic programming approach that characterizes, in an elegant and uniform way, the semantics of several AF-based frameworks. This result allows also to define the semantics for new AF-based frameworks, such as AFs with recursive attacks and recursive deductive supports. Under consideration for publication in Theory and Practice of Logic Programming.

We propose a novel ranking-based semantics for Dung-style argumentation frameworks with the help of conditional logics. Using an intuitive translation for an argumentation framework to generate conditionals, we can apply nonmonotonic inference systems to generate a ranking on possible worlds. With this ranking we construct a ranking for our arguments. With a small extension to this ranking-based semantics we already satisfy some desirable properties for a ranking over arguments.

The scariest thing about Artificial Intelligence is that we never know who the teacher is! If you're working on a Tech Startup, AI and Machine Learning are likely parts of your roadmap (and if it's not, then it should be). Artificial Intelligence (AI) is all around us. AI is there when you search for something on the Internet. AI helps us filter spam emails.

To understand the black-box characteristics of deep networks, counterfactual explanation that deduces not only the important features of an input space but also how those features should be modified to classify input as a target class has gained an increasing interest. The patterns that deep networks have learned from a training dataset can be grasped by observing the feature variation among various classes. However, current approaches perform the feature modification to increase the classification probability for the target class irrespective of the internal characteristics of deep networks. This often leads to unclear explanations that deviate from real-world data distributions. To address this problem, we propose a counterfactual explanation method that exploits the statistics learned from a training dataset. Especially, we gradually construct an explanation by iterating over masking and composition steps. The masking step aims to select an important feature from the input data to be classified as a target class. Meanwhile, the composition step aims to optimize the previously selected feature by ensuring that its output score is close to the logit space of the training data that are classified as the target class. Experimental results show that our method produces human-friendly interpretations on various classification datasets and verify that such interpretations can be achieved with fewer feature modification.

This post is about LSBoost, an Explainable'AI' algorithm which uses Gradient Boosted randomized networks for pattern recognition. In LSBoost, more specifically, the so called weak learners from LS_Boost are based on randomized neural networks' components and variants of Least Squares regression models. I've already presented some promising examples of use of LSBoost based on Ridge Regression weak learners. In mlsauce's version 0.7.1, the Lasso can also be used as an alternative ingredient to the weak learners. Here is a comparison of the regression coefficients obtained by using mlsauce's implementation of Ridge regression and the Lasso: The following example is about training set error vs testing set error, as a function of the regularization parameter, both for Ridge regression and Lasso-based weak learners.