We propose self-teaching networks to improve the generalization capacity of deep neural networks. The idea is to generate soft supervision labels using the output layer for training the lower layers of the network. During the network training, we seek an auxiliary loss that drives the lower layer to mimic the behavior of the output layer. The connection between the two network layers through the auxiliary loss can help the gradient flow, which works similar to the residual networks. Furthermore, the auxiliary loss also works as a regularizer, which improves the generalization capacity of the network. We evaluated the self-teaching network with deep recurrent neural networks on speech recognition tasks, where we trained the acoustic model using 30 thousand hours of data. We tested the acoustic model using data collected from 4 scenarios. We show that the self-teaching network can achieve consistent improvements and outperform existing methods such as label smoothing and confidence penalization.

-- In this paper we propose a deep neural network model with an encoder-decoder architecture that translates images of math formulas into their LaTeX markup sequences. The enc oder is a convolutional neural network (CNN) that transforms images into a group of feature maps. To better capture the spatia l relationships of math symbols, the feature maps are augmented with 2D positional encoding before being unfolded into a vector. The d ecoder is a stacked bidirectional long short-term memory (LSTM) model integrated with the soft attention mechanism, which works as a language model to translate the encoder output into a sequence of LaTeX tokens. The neural network is trained in two steps. The first step is token-level training using the Maximum-Like lihood Estimation (MLE) as the objective function. At comp letion of the token-level training, the sequence-level training objective function is employed to optimize the overall model based on the policy gradient algorithm from reinforcement learning. Our design a lso overcomes the exposure bias problem by closing the feedback l oop in the decoder during sequence-level training, i.e., feedi ng in the predicted token instead of the ground truth token at every time step. The model is trained and evaluated on the IM2LATEX-100K dataset and shows state-of-the-art performance on both sequence-based and image-based evaluation metrics. Math formulas often carry the most significant tech nical substances in many science, technology, engineering and math (STEM) fields. Being able to extract the math formulas from digital documents and translate them into markup la nguages is very useful for a wide range of information retriev al tasks. Portable Document Format (PDF) is the de facto standard publication format, which makes document distributi on very easy and reliable.

Non-Bayesian social learning theory provides a framework that models distributed inference for a group of agents interacting over a social network. In this framework, each agent iteratively forms and communicates beliefs about an unknown state of the world with their neighbors using a learning rule. Existing approaches assume agents have access to precise statistical models (in the form of likelihoods) for the state of the world. However in many situations, such models must be learned from finite data. We propose a social learning rule that takes into account uncertainty in the statistical models using second-order probabilities. Therefore, beliefs derived from uncertain models are sensitive to the amount of past evidence collected for each hypothesis. We characterize how well the hypotheses can be tested on a social network, as consistent or not with the state of the world. We explicitly show the dependency of the generated beliefs with respect to the amount of prior evidence. Moreover, as the amount of prior evidence goes to infinity, learning occurs and is consistent with traditional social learning theory.

Option discovery and skill acquisition frameworks are integral to the functioning of a Hierarchically organized Reinforcement learning agent. However, such techniques often yield a large number of options or skills, which can potentially be represented succinctly by filtering out any redundant information. Such a reduction can reduce the required computation while also improving the performance on a target task. In order to compress an array of option policies, we attempt to find a policy basis that accurately captures the set of all options. In this work, we propose Option Encoder, an auto-encoder based framework with intelligently constrained weights, that helps discover a collection of basis policies. The policy basis can be used as a proxy for the original set of skills in a suitable hierarchically organized framework. We demonstrate the efficacy of our method on a collection of grid-worlds and on the high-dimensional Fetch-Reach robotic manipulation task by evaluating the obtained policy basis on a set of downstream tasks.

We explore fixed-horizon temporal difference (TD) methods, reinforcement learning algorithms for a new kind of value function that predicts the sum of rewards over a $\textit{fixed}$ number of future time steps. To learn the value function for horizon $h$, these algorithms bootstrap from the value function for horizon $h-1$, or some shorter horizon. Because no value function bootstraps from itself, fixed-horizon methods are immune to the stability problems that plague other off-policy TD methods using function approximation (also known as "the deadly triad"). Although fixed-horizon methods require the storage of additional value functions, this gives the agent additional predictive power, while the added complexity can be substantially reduced via parallel updates, shared weights, and $n$-step bootstrapping. We show how to use fixed-horizon value functions to solve reinforcement learning problems competitively with methods such as Q-learning that learn conventional value functions. We also prove convergence of fixed-horizon temporal difference methods with linear and general function approximation. Taken together, our results establish fixed-horizon TD methods as a viable new way of avoiding the stability problems of the deadly triad.

A secret, pre-CIA field manual for sabotaging enemy organizations during WWII identified two ways of undermining an organization: physical damage (think: pulling out wires and destroying equipment) and human obstruction of processes (think: all the dysfunctional habits companies still struggle with today). When he shares the list of sabotaging tactics with executives today, HBS professor Stefan Thomke finds "that their reaction usually starts with laughter ('I see this in my company.'), Up until last Wednesday, the answer was probably "no." In fact, for years, even the most sophisticated AI systems have been unable to match the language and logic skills students are expected to have mastered before entering high school. But last week (just in time for back to school), the Allen Institute for Artificial Intelligence announced big news about its new system, Aristo -- it passed.

Increasingly, voice assistants from vendors such as Amazon, Apple, Google, Microsoft, and others are starting to find their way into myriad of devices, products, and tools used on a daily basis. While once we might have only interacted with conversational systems on our phones, dedicated desktop appliances, or desktop computers, we can now find conversational interfaces on a wide range of appliances and products from televisions to cars and even toaster ovens. Soon, any device we can interact with will have an audio conversational interface instead of buttons or screens to type or click. The dawn of the conversational computing age is here. However, are these devices intelligent enough to handle the wide range of queries that humans are posing?

When you want to learn a new technology for professional use, there are two mutually exclusive options, either you learn it yourself or you go for instructor based training. Self learning is least expensive but lot of time results in wasting time in finding right contents, setting up the environment, troubleshooting issues and may make you give up in the middle. Instructor based training can be expensive at times and need your time commitment. This course combines the best of both these options. The course is based on one of the most famous books in the field "Python Machine Learning (2nd Ed.)" by Sebastian Raschka and Vahid Mirjalili and provides you video tutorials on how to understand the AI/ML concepts from the books by providing out of box virtual machine with demo examples for each chapter in the book and complete preinstalled setup to execute the code.

At the Science 2019 Hofmann Lecture, Dina Katabi, PhD, will discuss her work and research.

The request was "rather strange," the director noted later in an email, but the voice was so lifelike that he felt he had no choice but to comply. The insurer, whose case was first reported by the Wall Street Journal, provided new details on the theft to The Washington Post on Wednesday, including an email from the employee tricked by what the insurer is referring to internally as "the false Johannes." Now being developed by a wide range of Silicon Valley titans and AI start-ups, such voice-synthesis software can copy the rhythms and intonations of a person's voice and be used to produce convincing speech. Tech giants such as Google and smaller firms such as the "ultrarealistic voice cloning" start-up Lyrebird have helped refine the resulting fakes and made the tools more widely available free for unlimited use. But the synthetic audio and AI-generated videos, known as "deepfakes," have fueled growing anxieties over how the new technologies can erode public trust, empower criminals and make traditional communication -- business deals, family phone calls, presidential campaigns -- that much more vulnerable to computerized manipulation.