As a grad student working on artificial intelligence, Mohamed Abdalla could probably walk into a number of well-paid industry jobs. Instead, he wants to draw attention to how Big Tech's big bucks may be warping the perspective of his field. Abdalla, who is finishing his PhD at the University of Toronto, has coauthored a paper highlighting the number of top AI researchers--including those who study the ethical challenges raised by the technology--who receive funding from tech companies. That can be a particular problem, he says, when corporate AI systems raise ethical issues, such as algorithmic bias, military use, or questions about the fairness and accuracy of face recognition programs. Abdalla found that more than half of tenure-track AI faculty at four prominent universities who disclose their funding sources have received some sort of backing from Big Tech.

I recently started a new newsletter focus on AI education. TheSequence is a no-BS( meaning no hype, no news etc) AI-focused newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. Natural language processing(NLP) has become the best known discipline in the deep learning space in rencet years. Part of that popularity have brought together an explosion of tools and frameworks such as Google Cloud, Azure LUIS, AWS Lex or Watson Assistant, NLP that have enable the implementation of simple NLP applications without requiring any deep learning knowledge.

The above figure is to show how a simpler version of SpinalNet can be converted to a single hidden layer NN. In Fig- 4(a), the first layer neurons are simplified by making them as linear functions. So, the first layer only takes the weighted sum of x1 to x5 inputs. Now, from the first layer, the output only goes to the corresponding neuron in the second layer. All cross-connections between neurons of two layers and the connections with the output layer are disconnected by assigning weight zero.

LinkedIn open-sources Generalized Deep Mixed Model (GDMix), a framework that makes training of AI personalization models more efficient. It supports deep learning models and so its an improvement over LinkedIn's previous release, Photon ML. GDMix trains two kinds of models: fixed effect and random effect models, used in search personalization and recommender systems. Usually, these models are challenging to teach in isolation. Still, GDMix accelerates the process by breaking the large models into a global model (fixed effect) and many small models (random effects) and then solving them separately.

Manifold learning is a fundamental problem in machine learning with numerous applications. Most of the existing methods directly learn the low-dimensional embedding of the data in some high-dimensional space, and usually lack the flexibility of being directly applicable to down-stream applications. In this paper, we propose the concept of implicit manifold learning, where manifold information is implicitly obtained by learning the associated heat kernel. A heat kernel is the solution of the corresponding heat equation, which describes how "heat" transfers on the manifold, thus containing ample geometric information of the manifold. We provide both practical algorithm and theoretical analysis of our framework. The learned heat kernel can be applied to various kernel-based machine learning models, including deep generative models (DGM) for data generation and Stein Variational Gradient Descent for Bayesian inference. Extensive experiments show that our framework can achieve state-of-the-art results compared to existing methods for the two tasks.

Slimmable neural networks provide a flexible trade-off front between prediction error and computational cost (such as the number of floating-point operations or FLOPs) with the same storage cost as a single model. They have been proposed recently for resource-constrained settings such as mobile devices. However, current slimmable neural networks use a single width-multiplier for all the layers to arrive at sub-networks with different performance profiles, which neglects that different layers affect the network's prediction accuracy differently and have different FLOP requirements. Hence, developing a principled approach for deciding width-multipliers across different layers could potentially improve the performance of slimmable networks. To allow for heterogeneous width-multipliers across different layers, we formulate the problem of optimizing slimmable networks from a multi-objective optimization lens, which leads to a novel algorithm for optimizing both the shared weights and the width-multipliers for the sub-networks. We perform extensive empirical analysis with 14 network and dataset combinations and find that less over-parameterized networks benefit more from a joint channel and weight optimization than extremely over-parameterized networks. Quantitatively, improvements up to 1.7% and 1% in top-1 accuracy on the ImageNet dataset can be attained for MobileNetV2 and MobileNetV3, respectively. Our results highlight the potential of optimizing the channel counts for different layers jointly with the weights for slimmable networks.

All datasets contain some biases, often unintentional, due to how they were acquired and annotated. These biases distort machine-learning models' performance, creating spurious correlations that the models can unfairly exploit, or, contrarily destroying clear correlations that the models could learn. With the popularity of deep learning models, automated skin lesion analysis is starting to play an essential role in the early detection of Melanoma. The ISIC Archive is one of the most used skin lesion sources to benchmark deep learning-based tools. Bissoto et al. experimented with different bounding-box based masks and showed that deep learning models could classify skin lesion images without clinically meaningful information in the input data. Their findings seem confounding since the ablated regions (random rectangular boxes) are not significant. The shape of the lesion is a crucial factor in the clinical characterization of a skin lesion. In that context, we performed a set of experiments that generate shape-preserving masks instead of rectangular bounding-box based masks. A deep learning model trained on these shape-preserving masked images does not outperform models trained on images without clinically meaningful information. That strongly suggests spurious correlations guiding the models. We propose use of general adversarial network (GAN) to mitigate the underlying bias.

Incorporating a so-called "momentum" dynamic in gradient descent methods is widely used in neural net training as it has been broadly observed that, at least empirically, it often leads to significantly faster convergence. At the same time, there are very few theoretical guarantees in the literature to explain this apparent acceleration effect. In this paper we show that Polyak's momentum, in combination with over-parameterization of the model, helps achieve faster convergence in training a one-layer ReLU network on $n$ examples. We show specifically that gradient descent with Polyak's momentum decreases the initial training error at a rate much faster than that of vanilla gradient descent. We provide a bound for a fixed sample size $n$, and we show that gradient descent with Polyak's momentum converges at an accelerated rate to a small error that is controllable by the number of neurons $m$. Prior work [DZPS19] showed that using vanilla gradient descent, and with a similar method of over-parameterization, the error decays as $(1-\kappa_n)^t$ after $t$ iterations, where $\kappa_n$ is a problem-specific parameter. Our result shows that with the appropriate choice of parameters one has a rate of $(1-\sqrt{\kappa_n})^t$. This work establishes that momentum does indeed speed up neural net training.

In game development, designing compelling visual assets that convey gameplay-relevant features requires time and experience. Recent image generation methods that create high-quality content could reduce development costs, but these approaches do not consider game mechanics. We propose a Convolutional Variational Autoencoder (CVAE) system to modify and generate new game visuals based on their gameplay relevance. We test this approach with Pok\'emon sprites and Pok\'emon type information, since types are one of the game's core mechanics and they directly impact the game's visuals. Our experimental results indicate that adopting a transfer learning approach can help to improve visual quality and stability over unseen data.

Reinforcement Learning (RL) agents typically learn memoryless policies---policies that only consider the last observation when selecting actions. Learning memoryless policies is efficient and optimal in fully observable environments. However, some form of memory is necessary when RL agents are faced with partial observability. In this paper, we study a lightweight approach to tackle partial observability in RL. We provide the agent with an external memory and additional actions to control what, if anything, is written to the memory. At every step, the current memory state is part of the agent's observation, and the agent selects a tuple of actions: one action that modifies the environment and another that modifies the memory. When the external memory is sufficiently expressive, optimal memoryless policies yield globally optimal solutions. Unfortunately, previous attempts to use external memory in the form of binary memory have produced poor results in practice. Here, we investigate alternative forms of memory in support of learning effective memoryless policies. Our novel forms of memory outperform binary and LSTM-based memory in well-established partially observable domains.