I will assume a basic understanding of neural networks and backpropagation. If you'd like to brush up, this lecture will give you the basics of neural networks and this one will explain how these principles are applied in modern deep learning systems. A working knowledge of Pytorch is required to understand the programming examples, but these can also be safely skipped. The fundamental operation of any transformer architecture is the self-attention operation. Self-attention is a sequence-to-sequence operation: a sequence of vectors goes in, and a sequence of vectors comes out. The vectors all have dimension \(k\). The weight \(w_{\rc{i}\gc{j}}\) is not a parameter, as in a normal neural net, but it is derived from a function over \(\x_\rc{i}\) and \(\x_\gc{j}\). A few other ingredients are needed for a complete transformer, which we'll discuss later, but this is the fundamental operation. More importantly, this is the only operation in the whole architecture that propagates information between vectors. Every other operation in the transformer is applied to each vector in the input sequence without interactions between vectors. Despite its simplicity, it's not immediately obvious why self-attention should work so well.

We've all seen the film where robots take over the world, with their mechanical bodies causing Hollywood-style screams from unsuspecting (or maybe very suspecting) victims. And, while these kinds of films let us live an alternate reality for an hour and a half, there's always that niggling thought at the backs of our minds telling us that this could actually happen in the not-too-distant future. In fact, the "father of AI", Alan Turing, was beavering away on it in the 1950s. He developed the Turing Test, which had a judge ask questions to a machine and a human. The judge would then have to decide who was the human and, if the computer could fool the judge at least half of the time, it was considered intelligent.

Truth be told, the answer to questions that start with "Can Artificial Intelligence..." will almost always be "yes" in the long term. However, is it really possible at some point in the future machines will be able to produce music indistinguishable from that of a human? Music is ultimately a set of patterns that humans find meaning in. Since it is closely related to our feelings and emotions, it is hard to imagine an "emotionless" robot composing the works of the great Chopin or Pink Floyd. Humans make music out of almost a need, a need to explain the "human experience."

Machine learning and artificial intelligence are playing a significant role in cybersecurity.

We describe the third edition of the CheckThat! Lab, which is part of the 2020 Cross-Language Evaluation Forum (CLEF). CheckThat! proposes four complementary tasks and a related task from previous lab editions, offered in English, Arabic, and Spanish. Task 1 asks to predict which tweets in a Twitter stream are worth fact-checking. Task 2 asks to determine whether a claim posted in a tweet can be verified using a set of previously fact-checked claims. Task 3 asks to retrieve text snippets from a given set of Web pages that would be useful for verifying a target tweet's claim. Task 4 asks to predict the veracity of a target tweet's claim using a set of Web pages and potentially useful snippets in them. Finally, the lab offers a fifth task that asks to predict the check-worthiness of the claims made in English political debates and speeches. CheckThat! features a full evaluation framework. The evaluation is carried out using mean average precision or precision at rank k for ranking tasks, and F1 for classification tasks.

Lack of understanding of the decisions made by model-based AI systems is an important barrier for their adoption. We examine counterfactual explanations as an alternative for explaining AI decisions. The counterfactual approach defines an explanation as a set of the system's data inputs that causally drives the decision (meaning that removing them changes the decision) and is irreducible (meaning that removing any subset of the inputs in the explanation does not change the decision). We generalize previous work on counterfactual explanations, resulting in a framework that (a) is model-agnostic, (b) can address features with arbitrary data types, (c) is able explain decisions made by complex AI systems that incorporate multiple models, and (d) is scalable to large numbers of features. We also propose a heuristic procedure to find the most useful explanations depending on the context. We contrast counterfactual explanations with another alternative: methods that explain model predictions by weighting features according to their importance (e.g., SHAP, LIME). This paper presents two fundamental reasons why explaining model predictions is not the same as explaining the decisions made using those predictions, suggesting we should carefully consider whether importance-weight explanations are well-suited to explain decisions made by AI systems. Specifically, we show that (1) features that have a large importance weight for a model prediction may not actually affect the corresponding decision, and (2) importance weights are insufficient to communicate whether and how features influence system decisions. We demonstrate this using several examples, including three detailed studies using real-world data that compare the counterfactual approach with SHAP and illustrate various conditions under which counterfactual explanations explain data-driven decisions better than feature importance weights.

Using one of your examples, for me the symbolic constraints for OCR of printed source code seem conceptually similar to how standard OCR systems implement the split between the character OCR model (which provides the probabilities of individual characters) and the language model (which provides the probabilities of long combined sequences of possible alternative characters) - any symbolic rules and constraints could be integrated straight into the language model, by adding an extra penalty to the likelihood of sequences that don't match some rule. The algorithms to effectively explore the solution space in this manner (often some form of beam search) already exist and would be already implemented and tested in an OCR system, so the symbolic rules would just change how the cost function is calculated.

Some tech CEOs run the show but don't really understand the nuts and bolts of the company's product. Riveiro is the CEO of Vilynx (pronounced "VEE-links"), a firm which uses artificial intelligence (AI) and machine learning to help media companies make video "smarter." To date, the Spanish company has received about $15 million in funding from European and North American venture capital and angel investors. Vilynx has offices in Barcelona, Palo Alto and New York City. This interview has been edited for brevity and clarity.

OTT platforms are becoming popular now than ever before. The competition between different OTT platforms has increased tremendously. Technologies like AI for OTT services can help businesses stay relevant in this cut-throat competition. According to a report, by the end of 2019, about 182.5 million people in the US will view content via OTT services. This represents more than 50% of the US population and provides huge business opportunities for OTT service providers.