When saying a word in your mind, your brain does not fully decouple the process of "sub-vocalizing" that word from speaking it, which can result in either minor or imperceptible movements of the mouth, tongue, larynx or other facial muscles.* The act of activating a muscle is not just a single "command" as we'd imagine in the digital world, but involves the repeated firing of multiple motor units (collections of muscle fibers and neuron terminals), at a rate of somewhere between 7–20 Hz, depending on the size and structure of the muscle. These firings will be providing us the electrical signal we are looking for, which we can read using an EMG sensor. To read the signals I used an OpenBCI board, technically designed for EEG, which I had on hand from some previous biofeedback experiments. EEG typically requires higher resolution, so if anything, this should help in picking up the weaker EMG signals we are looking for.

Global Artificial Intelligence in Retail Market was valued US$993.6 Mn in 2017 and is expected to reach US$8314 Mn by 2026, at a CAGR of 30.41% during a forecast period. The report is majorly segmented into types, technologies, solutions, services, deployment modes, applications, and region. Further, Artificial Intelligence in a retail market based on type includes online and offline retail. Technology segment is sub-segmented into machine learning and deep learning, Natural Language Processing, and others. Solution segment in the report comprises product recommendation & planning, customer relationship management, visual search, virtual assistant, price optimization, payment services management, supply chain management & demand planning, and others which include website and content optimization, space planning, and fraud detection.

Artificial Intelligence (AI) has provided remarkable capabilities and advances in image understanding, voice recognition, face recognition, pattern recognition, natural language processing, game planning, military applications, financial modeling, language translation, and search engine optimization. In medicine, deep learning is now one of the most powerful and promising tool of AI, which can enhance every stage of patient care --from research, omics data integration, combating antibiotic resistance bacteria, drug design and discovery to diagnosis and selection of appropriate therapy. It is also the key technology behind self-driving car. However, deep learning algorithms of AI have several inbuilt limitations. To utilize the full power of artificial intelligence, we need to know its strength and weakness and the ways to overcome those limitations in near future.

I recently stumbled upon this paper by the Google Brain team called "One model to learn them all". I just had to open it and take a look at the paper because of its great title (best ML paper title ever?). I quickly discovered that it is about a very fascinating and interesting idea. In this article I want to quickly summarize what I personally found the most interesting. Can we create a unified deep learning model to solve tasks across multiple domains?

I recently stumbled upon this paper by the Google Brain team called "One model to learn them all". I just had to open it and take a look at the paper because of its great title (best ML paper title ever?). I quickly discovered that it is about a very fascinating and interesting idea. In this article I want to quickly summarize what I personally found the most interesting. Can we create a unified deep learning model to solve tasks across multiple domains?

The #1 book that got the most votes is "Understanding Machine Learning: From Theory to Algorithms" by Shai Shalev-Shwartz and Shai Ben-David. The book was first published in 2014 by Cambridge University aiming for students who want to learn the basics of Machine Learning and be familiar with all the important algorithms in this field.

In semi-supervised learning, virtual adversarial training (VAT) approach is one of the most attractive method due to its intuitional simplicity and powerful performances. VAT finds a classifier which is robust to data perturbation toward the adversarial direction. In this study, we provide a fundamental explanation why VAT works well in semi-supervised learning case and propose new techniques which are simple but powerful to improve the VAT method. Especially we employ the idea of Bad GAN approach, which utilizes bad samples distributed on complement of the support of the input data, without any additional deep generative architectures. We generate bad samples of high-quality by use of the adversarial training used in VAT and also give theoretical explanations why the adversarial training is good at both generating bad samples. An advantage of our proposed method is to achieve the competitive performances compared with other recent studies with much fewer computations. We demonstrate advantages our method by various experiments with well known benchmark image datasets.

Ironies can not only express stronger emotions but also show a sense of humor. With the development of social media, ironies are widely used in public. Although many prior research studies have been conducted in irony detection, few studies focus on irony generation. The main challenges for irony generation are the lack of large-scale irony dataset and difficulties in modeling the ironic pattern. In this work, we first systematically define irony generation based on style transfer task. To address the lack of data, we make use of twitter and build a large-scale dataset. We also design a combination of rewards for reinforcement learning to control the generation of ironic sentences. Experimental results demonstrate the effectiveness of our model in terms of irony accuracy, sentiment preservation, and content preservation.

In this paper, we aim to provide elements to contribute to the discussion about the usefulness of deep CNNs with several filters to solve both within-subject and cross-subject classification for single-trial P300 detection. To that end, we present SepConv1D, a simple Convolutional Neural Network architecture consisting of a depthwise separable 1D convolutional block followed by a Sigmoid classification block. Additionally, we present a one-layer Fully-Connected Neural Network with two neurons in the hidden layer to show the unnecessary of having complex architectures to solve the problem under analysis. We compare their performances against CNN-based state-of-the-art architectures. The experiments did not show a statistically significant difference between their AUC. Moreover, SepConv1D has the lowest number of parameters of all by far. This is important because simpler, cheaper, faster and, thus, more portable devices can be built.

In this paper, we focus on the separability of classes with the cross-entropy loss function for classification problems by theoretically analyzing the intra-class distance and inter-class distance (i.e. the distance between any two points belonging to the same class and different classes, respectively) in the feature space, i.e. the space of representations learnt by neural networks. Specifically, we consider an arbitrary network architecture having a fully connected final layer with Softmax activation and trained using the cross-entropy loss. We derive expressions for the value and the distribution of the squared L2 norm of the product of a network dependent matrix and a random intra-class and inter-class distance vector (i.e. the vector between any two points belonging to the same class and different classes), respectively, in the learnt feature space (or the transformation of the original data) just before Softmax activation, as a function of the cross-entropy loss value. The main result of our analysis is the derivation of a lower bound for the probability with which the inter-class distance is more than the intra-class distance in this feature space, as a function of the loss value. We do so by leveraging some empirical statistical observations with mild assumptions and sound theoretical analysis. As per intuition, the probability with which the inter-class distance is more than the intra-class distance decreases as the loss value increases, i.e. the classes are better separated when the loss value is low. To the best of our knowledge, this is the first work of theoretical nature trying to explain the separability of classes in the feature space learnt by neural networks trained with the cross-entropy loss function.