Significant advances have been made in recent years on Natural Language Processing with machines surpassing human performance in many tasks, including but not limited to Question Answering. The majority of deep learning methods for Question Answering targets domains with large datasets and highly matured literature. The area of Nuclear and Atomic energy has largely remained unexplored in exploiting non-annotated data for driving industry viable applications. Due to lack of dataset, a new dataset was created from the 7000 research papers on nuclear domain. This paper contributes to research in understanding nuclear domain knowledge which is then evaluated on Nuclear Question Answering Dataset (NQuAD) created by nuclear domain experts as part of this research. NQuAD contains 612 questions developed on 181 paragraphs randomly selected from the IGCAR research paper corpus. In this paper, the Nuclear Bidirectional Encoder Representational Transformers (NukeBERT) is proposed, which incorporates a novel technique for building BERT vocabulary to make it suitable for tasks with less training data. The experiments evaluated on NQuAD revealed that NukeBERT was able to outperform BERT significantly, thus validating the adopted methodology. Training NukeBERT is computationally expensive and hence we will be open-sourcing the NukeBERT pretrained weights and NQuAD for fostering further research work in the nuclear domain.

The concentration of measure inequalities have received substantial attention in high-dimensional statistics and machine learning [1]. While concentration inequalities are well-understood for subGaussian and subexponential random variables, in many application areas, such as signal processing [2] and machine learning [3] we need concentration results for sums of random variables with heavier tails. The standard technique, i.e. finding upper bounds for the moment generating function (MGF), clearly fails for heavy-tailed distributions whose moment generating functions do not exist. Furthermore, other techniques, such as Chebyshev's inequality, are incapable of obtaining sharp results. The goal of this paper is to show that under quite general conditions on the tail a simple truncation argument can not only help us use the standard MGF argument for heavy-tailed random variables, but is also capable of obtaining sharp concentration results.

This paper describes the formulation and experimental testing of a novel method for the estimation and approximation of submanifold models of animal motion. It is assumed that the animal motion is supported on a configuration manifold $Q$ that is a smooth, connected, regularly embedded Riemannian submanifold of Euclidean space $X\approx \mathbb{R}^d$ for some $d>0$, and that the manifold $Q$ is homeomorphic to a known smooth, Riemannian manifold $S$. Estimation of the manifold is achieved by finding an unknown mapping $\gamma:S\rightarrow Q\subset X$ that maps the manifold $S$ into $Q$. The overall problem is cast as a distribution-free learning problem over the manifold of measurements $\mathbb{Z}=S\times X$. That is, it is assumed that experiments generate a finite sets $\{(s_i,x_i)\}_{i=1}^m\subset \mathbb{Z}^m$ of samples that are generated according to an unknown probability density $\mu$ on $\mathbb{Z}$. This paper derives approximations $\gamma_{n,m}$ of $\gamma$ that are based on the $m$ samples and are contained in an $N(n)$ dimensional space of approximants. The paper defines sufficient conditions that shows that the rates of convergence in $L^2_\mu(S)$ correspond to those known for classical distribution-free learning theory over Euclidean space. Specifically, the paper derives sufficient conditions that guarantee rates of convergence that have the form $$\mathbb{E} \left (\|\gamma_\mu^j-\gamma_{n,m}^j\|_{L^2_\mu(S)}^2\right )\leq C_1 N(n)^{-r} + C_2 \frac{N(n)\log(N(n))}{m}$$for constants $C_1,C_2$ with $\gamma_\mu:=\{\gamma^1_\mu,\ldots,\gamma^d_\mu\}$ the regressor function $\gamma_\mu:S\rightarrow Q\subset X$ and $\gamma_{n,m}:=\{\gamma^1_{n,j},\ldots,\gamma^d_{n,m}\}$.

This work demonstrates that, for a simple case of 10 randomly positioned atoms, a neural network can be trained to infer atomic coordinates from Patterson maps. The network was trained entirely on synthetic data. For the training set, the network outputs were 3D maps of randomly positioned atoms. From each output map, a Patterson map was generated and used as input to the network. The network generalized to cases not in the test set, inferring atom positions from Patterson maps. A key finding in this work is that the Patterson maps presented to the network input during training must uniquely describe the atomic coordinates they are paired with on the network output or the network will not train and it will not generalize. The network cannot train on conflicting data. Avoiding conflicts is handled in 3 ways: 1. Patterson maps are invariant to translation. To remove this degree of freedom, output maps are centered on the average of their atom positions. 2. Patterson maps are invariant to centrosymmetric inversion. This conflict is removed by presenting the network output with both the atoms used to make the Patterson Map and their centrosymmetry-related counterparts simultaneously. 3. The Patterson map does not uniquely describe a set of coordinates because the origin for each vector in the Patterson map is ambiguous. By adding empty space around the atoms in the output map, this ambiguity is removed. Forcing output atoms to be closer than half the output box edge dimension means the origin of each peak in the Patterson map must be the origin to which it is closest.

Federated learning is a machine learning setting where a set of edge devices iteratively train a model under the orchestration of a central server, while keeping all data locally on edge devices. In each iteration of federated learning, edge devices perform computation with their local data, and the local computation results are then uploaded to the server for model update. During this process, the challenges of privacy leakage and communication overhead arise due to the extensive information exchange between edge devices and the server. In this paper, we develop CPFed, a communication-efficient and privacy-preserving federated learning method, to solve the above challenges. CPFed integrates three key components: (1) periodic averaging where local computation results at edge devices are only periodically averaged at the server; (2) Gaussian mechanism where edge devices randomly perturb their local computation results before sending the results to the server; and (3) secure aggregation where the perturbed local computation results are homomorphically encrypted before being sent to the server. CPFed can address both the communication efficiency and privacy leakage challenges in federated learning while achieving high model accuracy. We provide an end-to-end privacy guarantee of CPFed and analyze its theoretical convergence rates for both convex and non-convex models. Through extensive numerical experiments on real-world datasets, we demonstrate the effectiveness and efficiency of our proposed method.

We propose a novel regularization-based continual learning method, dubbed as Adaptive Group Sparsity based Continual Learning (AGS-CL), using two group sparsity-based penalties. Our method selectively employs the two penalties when learning each node based its the importance, which is adaptively updated after learning each new task. By utilizing the proximal gradient descent method for learning, the exact sparsity and freezing of the model is guaranteed, and thus, the learner can explicitly control the model capacity as the learning continues. Furthermore, as a critical detail, we re-initialize the weights associated with unimportant nodes after learning each task in order to prevent the negative transfer that causes the catastrophic forgetting and facilitate efficient learning of new tasks. Throughout the extensive experimental results, we show that our AGS-CL uses much less additional memory space for storing the regularization parameters, and it significantly outperforms several state-of-the-art baselines on representative continual learning benchmarks for both supervised and reinforcement learning tasks.

Oversmoothing has been assumed to be the major cause of performance drop in deep graph convolutional networks (GCNs). The evidence is usually derived from Simple Graph Convolution (SGC), a linear variant of GCNs. In this paper, we revisit graph node classification from an optimization perspective and argue that GCNs can actually learn anti-oversmoothing, whereas overfitting is the real obstacle in deep GCNs. This work interprets GCNs and SGCs as two-step optimization problems and provides the reason why deep SGC suffers from oversmoothing but deep GCNs does not. Our conclusion is compatible with the previous understanding of SGC, but we clarify why the same reasoning does not apply to GCNs. Based on our formulation, we provide more insights into the convolution operator and further propose a mean-subtraction trick to accelerate the training of deep GCNs. We verify our theory and propositions on three graph benchmarks. The experiments show that (i) in GCN, overfitting leads to the performance drop and oversmoothing does not exist even model goes to very deep (100 layers); (ii) mean-subtraction speeds up the model convergence as well as retains the same expressive power; (iii) the weight of neighbor averaging (1 is the common setting) does not significantly affect the model performance once it is above the threshold ( 0.5).

Generative adversarial networks (GANs) are a class of generative models, known for producing accurate samples. The key feature of GANs is that there are two antagonistic neural networks: the generator and the discriminator. The main bottleneck for their implementation is that the neural networks are very hard to train. One way to improve their performance is to design reliable algorithms for the adversarial process. Since the training can be cast as a stochastic Nash equilibrium problem, we rewrite it as a variational inequality and introduce an algorithm to compute an approximate solution. Specifically, we propose a stochastic relaxed forward-backward algorithm for GANs. We prove that when the pseudogradient mapping of the game is monotone, we have convergence to an exact solution or in a neighbourhood of it.

In this paper, we propose a novel model for time series prediction in which difference-attention LSTM model and error-correction LSTM model are respectively employed and combined in a cascade way. While difference-attention LSTM model introduces a difference feature to perform attention in traditional LSTM to focus on the obvious changes in time series. Error-correction LSTM model refines the prediction error of difference-attention LSTM model to further improve the prediction accuracy. Finally, we design a training strategy to jointly train the both models simultaneously. With additional difference features and new principle learning framework, our model can improve the prediction accuracy in time series. Experiments on various time series are conducted to demonstrate the effectiveness of our method.

We present a new class of stochastic, geometrically-driven optimization algorithms on the orthogonal group $O(d)$ and naturally reductive homogeneous manifolds obtained from the action of the rotation group $SO(d)$. We theoretically and experimentally demonstrate that our methods can be applied in various fields of machine learning including deep, convolutional and recurrent neural networks, reinforcement learning, normalizing flows and metric learning. We show an intriguing connection between efficient stochastic optimization on the orthogonal group and graph theory (e.g. matching problem, partition functions over graphs, graph-coloring). We leverage the theory of Lie groups and provide theoretical results for the designed class of algorithms. We demonstrate broad applicability of our methods by showing strong performance on the seemingly unrelated tasks of learning world models to obtain stable policies for the most difficult $\mathrm{Humanoid}$ agent from $\mathrm{OpenAI}$ $\mathrm{Gym}$ and improving convolutional neural networks.