Mean field theory has been successfully used to analyze deep neura l networks (DNN) in the infinite size limit. Given the finite size of realistic D NN, we utilize the large deviation theory and path integral analysis to study the deviation of functions represented by DNN from their typical mean field solution s. The parameter perturbations investigated include weight sparsification (dilution) a nd binarization, which are commonly used in model simplification, for both ReLU and sign activation functions. We find that random networks with ReLU activation are m ore robust to parameter perturbations with respect to their counterparts wit h sign activation, which arguably is reflected in the simplicity of the functions they generate . Keywords: large deviation theory, path integral, deep neural networks, fu nction sensitivity 1. Introduction Learning machines realized by deep neural networks (DNN) have ac hieved impressive success in performing various machine learning tasks, such as spee ch recognition, image classification and natural language processing [1].

Hidden Markov model (HMM) has been successfully used for sequential data modeling problems. In this work, we propose to power the modeling capacity of HMM by bringing in neural network based generative models. The proposed model is termed as GenHMM. In the proposed GenHMM, each HMM hidden state is associated with a neural network based generative model that has tractability of exact likel i-hood and provides efficient likelihood computation. A generative model in GenHMM consists of mixture of generators that are realized by flow models. A learning algorithm for GenHMM is proposed in expectation-maximization framework. The convergence of the learning GenHMM is analyzed. We demonstrate the efficiency of GenHMM by classification tasks on practical sequential data.

Recent work in signal propagation theory has shown that dropout limits the depth to which information can propagate through a neural network. In this paper, we investigate the effect of initialisation on training speed and generalisation for ReLU networks within this depth limit. We ask the following research question: given that critical initialisation is crucial for training at large depth, if dropout limits the depth at which networks are trainable, does initialising critically still matter? We conduct a large-scale controlled experiment, and perform a statistical analysis of over $12000$ trained networks. We find that (1) trainable networks show no statistically significant difference in performance over a wide range of non-critical initialisations; (2) for initialisations that show a statistically significant difference, the net effect on performance is small; (3) only extreme initialisations (very small or very large) perform worse than criticality. These findings also apply to standard ReLU networks of moderate depth as a special case of zero dropout. Our results therefore suggest that, in the shallow-to-moderate depth setting, critical initialisation provides zero performance gains when compared to off-critical initialisations and that searching for off-critical initialisations that might improve training speed or generalisation, is likely to be a fruitless endeavour.

We investigate the sample complexity of networks with bounds on the magnitude of its weights. In particular, we consider the class \[ H=\left\{W_t\circ\rho\circ \ldots\circ\rho\circ W_{1} :W_1,\ldots,W_{t-1}\in M_{d, d}, W_t\in M_{1,d}\right\} \] where the spectral norm of each $W_i$ is bounded by $O(1)$, the Frobenius norm is bounded by $R$, and $\rho$ is the sigmoid function $\frac{e^x}{1+e^x}$ or the smoothened ReLU function $ \ln (1+e^x)$. We show that for any depth $t$, if the inputs are in $[-1,1]^d$, the sample complexity of $H$ is $\tilde O\left(\frac{dR^2}{\epsilon^2}\right)$. This bound is optimal up to log-factors, and substantially improves over the previous state of the art of $\tilde O\left(\frac{d^2R^2}{\epsilon^2}\right)$. We furthermore show that this bound remains valid if instead of considering the magnitude of the $W_i$'s, we consider the magnitude of $W_i - W_i^0$, where $W_i^0$ are some reference matrices, with spectral norm of $O(1)$. By taking the $W_i^0$ to be the matrices at the onset of the training process, we get sample complexity bounds that are sub-linear in the number of parameters, in many typical regimes of parameters. To establish our results we develop a new technique to analyze the sample complexity of families $H$ of predictors. We start by defining a new notion of a randomized approximate description of functions $f:X\to\mathbb{R}^d$. We then show that if there is a way to approximately describe functions in a class $H$ using $d$ bits, then $d/\epsilon^2$ examples suffices to guarantee uniform convergence. Namely, that the empirical loss of all the functions in the class is $\epsilon$-close to the true loss. Finally, we develop a set of tools for calculating the approximate description length of classes of functions that can be presented as a composition of linear function classes and non-linear functions.

It is well established that temporal organization is critical to memory, and that the ability to temporally organize information is fundamental to many perceptual, cognitive, and motor processes. While our understanding of how the brain processes the spatial context of memories has advanced considerably, our understanding of their temporal organization lags far behind. In this paper, we propose a new approach for elucidating the neural basis of complex behaviors and temporal organization of memories. More specifically, we focus on neural decoding - the prediction of behavioral or experimental conditions based on observed neural data. In general, this is a challenging classification problem, which is of immense interest in neuroscience. Our goal is to develop a new framework that not only improves the overall accuracy of decoding, but also provides a clear latent representation of the decoding process. To accomplish this, our approach uses a Variational Auto-encoder (VAE) model with a diversity-encouraging prior based on determinantal point processes (DPP) to improve latent representation learning by avoiding redundancy in the latent space. We apply our method to data collected from a novel rat experiment that involves presenting repeated sequences of odors at a single port and testing the rats' ability to identify each odor. We show that our method leads to substantially higher accuracy rate for neural decoding and allows to discover novel biological phenomena by providing a clear latent representation of the decoding process.

The increasing popularity of internet, wireless technologies and mobile devices has led to the birth of mass connectivity and online interaction through Online Social Networks (OSNs) and similar environments. OSN reflects a social structure consist of a set of individuals and different types of ties like connections, relationships, interactions etc among them and helps its users to connect with their friends and common interest groups, share views and to pass information. Now days the u sers choose OSN sites as a most preferred place for sharing their updates, different views, posting photographs and would like to make it available for others for viewing, rating a nd making comments. Th e current paper aims to explore and analyze the association between the objects (like photographs, posts etc) and its viewers (friends, acquaintances etc) for a given user and to find activity relationship amo ng them by using the TF - I DF scheme of Vector Space Model. After vectorization the vector data has been presented through a weighted graph with various properties .

Most neural network designs for FPGAs are inflexible. In this paper, we propose a flexible VHDL structure that would allow any neural network to be implemented on multiple FPGAs. Moreover, the VHDL structure allows for testing as well as training multiple neural networks. The VHDL design consists of multiple processor groups. There are two types of processor groups: Mini Vector Machine Processor Group and Activation Processor Group. Each processor group consists of individual Mini Vector Machines and Activation Processor. The Mini Vector Machines apply vector operations to the data, while the Activation Processors apply activation functions to the data. A ring buffer was implemented to connect the various processor groups.

We present a taxonomy of research on Machine Learning (ML) applied to enhance simulations together with a catalog of some activities. We cover eight patterns for the link of ML to the simulations or systems plus three algorithmic areas: particle dynamics, agent-based models and partial differential equations. The patterns are further divided into three action areas: Improving simulation with Configurations and Integration of Data, Learn Structure, Theory and Model for Simulation, and Learn to make Surrogates.

A BSTRACT Owing to their ability to both effectively integrate information over long time horizons and scale to massive amounts of data, self-attention architectures have recently shown breakthrough success in natural language processing (NLP), achieving state-of-the-art results in domains such as language modeling and machine translation. Harnessing the transformer's ability to process long time horizons of information could provide a similar performance boost in partially observable reinforcement learning (RL) domains, but the large-scale transformers used in NLP have yet to be successfully applied to the RL setting. In this work we demonstrate that the standard transformer architecture is difficult to optimize, which was previously observed in the supervised learning setting but becomes especially pronounced with RL objectives. We propose architectural modifications that substantially improve the stability and learning speed of the original Transformer and XL variant. The proposed architecture, the Gated Transformer-XL (GTrXL), surpasses LSTMs on challenging memory environments and achieves state-of-the-art results on the multi-task DMLab-30 benchmark suite, exceeding the performance of an external memory architecture. We show that the GTrXL, trained using the same losses, has stability and performance that consistently matches or exceeds a competitive LSTM baseline, including on more reactive tasks where memory is less critical. GTrXL offers an easy-to-train, simple-to-implement but substantially more expressive architectural alternative to the standard multi-layer LSTM ubiquitously used for RL agents in partially observable environments. Recent work has empirically validated these claims, demonstrating that self-attention architectures can provide significant gains in performance over the more traditional recurrent architectures such as the LSTM (Dai et al., 2019; Radford et al., 2019; Devlin et al., 2019; Y ang et al., 2019). The repeated success of the transformer architecture in domains where sequential information processing is critical to performance makes it an ideal candidate for partially observable RL problems, where episodes can extend to thousands of steps and the critical observations for any decision often DeepMind and Machine Learning Department, Carnegie Mellon University. 1 arXiv:1910.06764v1 Y et, the RL literature is dominated by the use of LSTMs as the main mechanism for providing memory to the agent (Espeholt et al., 2018; Kapturowski et al., 2019; Mnih et al., 2016).

A BSTRACT Neural inductive program synthesis is a task generating instructions that can produce desired outputs from given inputs. In this paper, we focus on the generation of a chunk of assembly code that can be executed to match a state change inside the CPU and RAM. We develop a neural program synthesis algorithm, AutoAssem-blet, learned via self-learning reinforcement learning that explores the large code space efficiently. Policy networks and value networks are learned to reduce the breadth and depth of the Monte Carlo Tree Search, resulting in better synthesis performance. We also propose an effective multi-entropy policy sampling technique to alleviate online update correlations. We apply AutoAssemblet to basic programming tasks and show significant higher success rates compared to several competing baselines. Much progress has been made in the field with the development of methods along the vein of neural program synthesis (Parisotto et al., 2016; Balog et al., 2017; Bunel et al., 2018; Hayati et al., 2018; Desai et al., 2016; Yin & Neubig, 2017; Kant, 2018). Neural program synthesis models build on the top of neural network architectures to synthesize human-readable programs that match desired executions.