This code is an implementation of "Masked AutoEncoder for Density Estimation" by Germain et al., 2015. The core idea is that you can turn an auto-encoder into an autoregressive density model just by appropriately masking the connections in the MLP, ordering the input dimensions in some way and making sure that all outputs only depend on inputs earlier in the list. Like other autoregressive models (char-rnn, pixel cnns, etc), evaluating the likelihood is very cheap (a single forward pass), but sampling is linear in the number of dimensions. The authors of the paper also published code here, but it's a bit wordy, sprawling and in Theano. We can then simultaneously train a larger model ensemble (with weight sharing in the one MLP) and average over all of the models at test time.

We present a novel method for training deep neural network amenable to inference in low-precision arithmetic with quantized weights and activations. The training is performed in full precision with random noise injection emulating quantization noise. In order to circumvent the need to simulate realistic quantization noise distributions, the weight and the activation distributions are uniformized by a non-linear transformation, and uniform noise is injected. An inverse transformation is then applied. This procedure emulates a non-uniform k-quantile quantizer at inference time, which is shown to achieve state-of-the-art results for training low-precision networks on CIFAR-10 and ImageNet-1K datasets. In particular, we observe no degradation in accuracy for MobileNet and ResNet-18 on ImageNet with as low as 2-bit quantization of the activations and minimal degradation for as little as 4 bits for the weights.

The demands of UQ in computer prediction, a problem we believe to be NP-Hard, cannot be met on our current HPC technology path. We see that cognitive computing, defined through the technology convergence of AI, Big Data and HPC is an essential next step. With vendor technology decisions being made now and in the next few years in AI and HPC, it is urgent that broad classes of HW and SW are explored to best leverage commercial technology roadmaps. To that end, we are using precision medicine data as a force multiplier and accelerator. This rich, complex, unstructured, heterogeneous, curated, massive data is likely the richest class of data to work on today and brings with it unique partnerships that buys down risk in exploring the many splintered paths forward each with their own tough challenges and also shares costs.

We present a new approach to learning for planning, where knowledge acquired while solving a given set of planning problems is used to plan faster in related, but new problem instances. We show that a deep neural network can be used to learn and represent a \emph{generalized reactive policy} (GRP) that maps a problem instance and a state to an action, and that the learned GRPs efficiently solve large classes of challenging problem instances. In contrast to prior efforts in this direction, our approach significantly reduces the dependence of learning on handcrafted domain knowledge or feature selection. Instead, the GRP is trained from scratch using a set of successful execution traces. We show that our approach can also be used to automatically learn a heuristic function that can be used in directed search algorithms. We evaluate our approach using an extensive suite of experiments on two challenging planning problem domains and show that our approach facilitates learning complex decision making policies and powerful heuristic functions with minimal human input. Videos of our results are available at goo.gl/Hpy4e3.

In this work, we study the credit assignment problem in reward augmented maximum likelihood (RAML) learning, and establish a theoretical equivalence between the token-level counterpart of RAML and the entropy regularized reinforcement learning. Inspired by the connection, we propose two sequence prediction algorithms, one extending RAML with fine-grained credit assignment and the other improving Actor-Critic with a systematic entropy regularization. On two benchmark datasets, we show the proposed algorithms outperform RAML and Actor-Critic respectively, providing new alternatives to sequence prediction.

Recent research has shown that deep ensemble for forest can achieve a huge increase in classification accuracy compared with the general ensemble learning method. Especially when there are only few training data. In this paper, we decide to take full advantage of this observation and introduce the Dense Adaptive Cascade Forest (daForest), which has better performance than the original one named Cascade Forest. And it is particularly noteworthy that daForest has a powerful ability to handle high-dimensional sparse data without any preprocessing on raw data like PCA or any other dimensional reduction methods. Our model is distinguished by three major features: the first feature is the combination of the SAMME.R boosting algorithm in the model, boosting gives the model the ability to continuously improve as the number of layer increases, which is not possible in stacking model or plain cascade forest. The second feature is our model connects each layer to its subsequent layers in a feed-forward fashion, to some extent this structure enhances the ability of the model to resist degeneration. When number of layers goes up, accuracy of model goes up a little in the first few layers then drop down quickly, we call this phenomenon degeneration in training stacking model. The third feature is that we add a hyper-parameter optimization layer before the first classification layer in the proposed deep model, which can search for the optimal hyper-parameter and set up the model in a brief period and nearly halve the training time without having too much impact on the final performance. Experimental results show that daForest performs particularly well on both high-dimensional low-order features and low-dimensional high-order features, and in some cases, even better than neural networks and achieves state-of-the-art results.

Rogue is a famous dungeon-crawling video-game of the 80ies, the ancestor of its gender. Rogue-like games are known for the necessity to explore partially observable and always different randomly-generated labyrinths, preventing any form of level replay. As such, they serve as a very natural and challenging task for reinforcement learning, requiring the acquisition of complex, non-reactive behaviors involving memory and planning. In this article we show how, exploiting a version of A3C partitioned on different situations, the agent is able to reach the stairs and descend to the next level in 98% of cases.

In the figure above, two numbers are supposed to be added. Those numbers are stored in two variables, a and b. The two values are flowing through the graph and arrive at the square node, where they are being added. The result of the addition is stored into another variable, c. Actually, a, b and c can be considered as placeholders. Any numbers that are fed into a and b get added and are stored into c. This is exactly how TensorFlow works. The user defines an abstract representation of the model (neural network) through placeholders and variables. Afterwards, the placeholders get "filled" with real data and the actual computations take place.

Creating intuitive and engaging experiences for users is a critical goal for companies of all sizes, and it's a process driven by quick cycles of prototyping, designing, and user testing. Large corporations like Facebook have the bandwidth to dedicate entire teams to the design process, which can take several weeks and involve multiple stakeholders; small businesses don't have these resources, and their user interfaces may suffer as a result. My goal at Insight was to use modern deep learning algorithms to significantly streamline the design workflow and empower any business to quickly create and test webpages. The length of the development cycle can quickly turn into a bottleneck, and companies like Airbnb have started to use machine learning to make this process more efficient. Though promising as an example of machine-assisted design, it's unclear how much of this model is being fully trained end-to-end, and how much relies on hand-crafted image features.

Learning in models with discrete latent variables is challenging due to high-variance gradient estimators. Previous approaches either produced high-variance, unbiased gradients or low-variance, biased gradients. REBAR uses control variates and the reparameterization trick to get the best of both: low-variance, unbiased gradients that result in faster convergence to a better result. "We describe a new family of approaches for imagination-based planning...We also introduce architectures which provide new ways for agents to learn and construct plans to maximise the efficiency of a task. These architectures are efficient, robust to complex and imperfect models, and can adopt flexible strategies for exploiting their imagination. The agents we introduce benefit from an'imagination encoder'- a neural network which learns to extract any information useful for the agent's future decisions, but ignore that which is not relevant."