Polygon meshes are an efficient representation of 3D geometry, and are of central importance in computer graphics, robotics and games development. Existing learning-based approaches have avoided the challenges of working with 3D meshes, instead using alternative object representations that are more compatible with neural architectures and training approaches. We present an approach which models the mesh directly, predicting mesh vertices and faces sequentially using a Transformer-based architecture. Our model can condition on a range of inputs, including object classes, voxels, and images, and because the model is probabilistic it can produce samples that capture uncertainty in ambiguous scenarios. We show that the model is capable of producing high-quality, usable meshes, and establish log-likelihood benchmarks for the mesh-modelling task. We also evaluate the conditional models on surface reconstruction metrics against alternative methods, and demonstrate competitive performance despite not training directly on this task.

On-chip edge intelligence has necessitated the exploration of algorithmic techniques to reduce the compute requirements of current machine learning frameworks. This work aims to bridge the recent algorithmic progress in training Binary Neural Networks and Spiking Neural Networks - both of which are driven by the same motivation and yet synergies between the two have not been fully explored. We show that training Spiking Neural Networks in the extreme quantization regime results in near full precision accuracies on large-scale datasets like CIFAR-$100$ and ImageNet. An important implication of this work is that Binary Spiking Neural Networks can be enabled by "In-Memory" hardware accelerators catered for Binary Neural Networks without suffering any accuracy degradation due to binarization. We utilize standard training techniques for non-spiking networks to generate our spiking networks by conversion process and also perform an extensive empirical analysis and explore simple design-time and run-time optimization techniques for reducing inference latency of spiking networks (both for binary and full-precision models) by an order of magnitude over prior work.

For time series classification task using 1D-CNN, the selection of kernel size is critically important to ensure the model can capture the right scale salient signal from a long time-series. Most of the existing work on 1D-CNN treats the kernel size as a hyper-parameter and tries to find the proper kernel size through a grid search which is time-consuming and is inefficient. This paper theoretically analyses how kernel size impacts the performance of 1D-CNN. Considering the importance of kernel size, we propose a novel Omni-Scale 1D-CNN (OS-CNN) architecture to capture the proper kernel size during the model learning period. A specific design for kernel size configuration is developed which enables us to assemble very few kernel-size options to represent more receptive fields. The proposed OS-CNN method is evaluated using the UCR archive with 85 datasets. The experiment results demonstrate that our method is a stronger baseline in multiple performance indicators, including the critical difference diagram, counts of wins, and average accuracy. We also published the experimental source codes at GitHub (https://github.com/Wensi-Tang/OS-CNN/).

In this paper, we consider the source of Deep Reinforcement Learning (DRL)'s sample complexity, asking how much derives from the requirement of learning useful representations of environment states and how much is due to the sample complexity of learning a policy. While for DRL agents, the distinction between representation and policy may not be clear, we seek new insight through a set of transfer learning experiments. In each experiment, we retain some fraction of layers trained on either the same game or a related game, comparing the benefits of transfer learning to learning a policy from scratch. Interestingly, we find that benefits due to transfer are highly variable in general and non-symmetric across pairs of tasks. Our experiments suggest that perhaps transfer from simpler environments can boost performance on more complex downstream tasks and that the requirements of learning a useful representation can range from negligible to the majority of the sample complexity, based on the environment. Furthermore, we find that fine-tuning generally outperforms training with the transferred layers frozen, confirming an insight first noted in the classification setting.

The study of model bias and variance with respect to decision boundaries is critically important in supervised classification. There is generally a tradeoff between the two, as fine-tuning of the decision boundary of a classification model to accommodate more boundary training samples (i.e., higher model complexity) may improve training accuracy (i.e., lower bias) but hurt generalization against unseen data (i.e., higher variance). By focusing on just classification boundary fine-tuning and model complexity, it is difficult to reduce both bias and variance. To overcome this dilemma, we take a different perspective and investigate a new approach to handle inaccuracy and uncertainty in the training data labels, which are inevitable in many applications where labels are conceptual and labeling is performed by human annotators. The process of classification can be undermined by uncertainty in the labels of the training data; extending a boundary to accommodate an inaccurately labeled point will increase both bias and variance. Our novel method can reduce both bias and variance by estimating the pointwise label uncertainty of the training set and accordingly adjusting the training sample weights such that those samples with high uncertainty are weighted down and those with low uncertainty are weighted up. In this way, uncertain samples have a smaller contribution to the objective function of the model's learning algorithm and exert less pull on the decision boundary. In a real-world physical activity recognition case study, the data presents many labeling challenges, and we show that this new approach improves model performance and reduces model variance.

The enormous inference cost of deep neural networks can be scaled down by network compression. Pruning is one of the predominant approaches used for deep network compression. However, existing pruning techniques have one or more of the following limitations: 1) Additional energy cost on top of the compute heavy training stage due to pruning and fine-tuning stages, 2) Layer-wise pruning based on the statistics of a particular, ignoring the effect of error propagation in the network, 3) Lack of an efficient estimate for determining the important channels globally, 4) Unstructured pruning requires specialized hardware for effective use. To address all the above issues, we present a simple-yet-effective gradual channel pruning while training methodology using a novel data driven metric referred as Feature relevance score. The proposed technique gets rid of the additional retraining cycles by pruning least important channels in a structured fashion at fixed intervals during the actual training phase. Feature relevance scores help in efficiently evaluating the contribution of each channel towards the discriminative power of the network. We demonstrate the effectiveness of the proposed methodology on architectures such as VGG and ResNet using datasets such as CIFAR-10, CIFAR-100 and ImageNet, and successfully achieve significant model compression while trading off less than $1\%$ accuracy. Notably on CIFAR-10 dataset trained on ResNet-110, our approach achieves $2.4\times$ compression and a $56\%$ reduction in FLOPs with an accuracy drop of $0.01\%$ compared to the unpruned network.

In spite of several notable efforts, explaining the generalization of deterministic deep nets, e.g., ReLU-nets, has remained challenging. Existing approaches usually need to bound the Lipschitz constant of such deep nets but such bounds have been shown to increase substantially with the number of training samples yielding vacuous generalization bounds [Nagarajan and Kolter, 2019a]. In this paper, we present new de-randomized PAC-Bayes margin bounds for deterministic non-convex and non-smooth predictors, e.g., ReLU-nets. The bounds depend on a trade-off between the $L_2$-norm of the weights and the effective curvature (`flatness') of the predictor, avoids any dependency on the Lipschitz constant, and yield meaningful (decreasing) bounds with increase in training set size. Our analysis first develops a de-randomization argument for non-convex but smooth predictors, e.g., linear deep networks (LDNs). We then consider non-smooth predictors which for any given input realize as a smooth predictor, e.g., ReLU-nets become some LDN for a given input, but the realized smooth predictor can be different for different inputs. For such non-smooth predictors, we introduce a new PAC-Bayes analysis that maintains distributions over the structure as well as parameters of smooth predictors, e.g., LDNs corresponding to ReLU-nets, which after de-randomization yields a bound for the deterministic non-smooth predictor. We present empirical results to illustrate the efficacy of our bounds over changing training set size and randomness in labels.

Autoregressive models (ARMs) currently hold state-of-the-art performance in likelihood-based modeling of image and audio data. Generally, neural network based ARMs are designed to allow fast inference, but sampling from these models is impractically slow. In this paper, we introduce the predictive sampling algorithm: a procedure that exploits the fast inference property of ARMs in order to speed up sampling, while keeping the model intact. We propose two variations of predictive sampling, namely sampling with ARM fixed-point iteration and learned forecasting modules. Their effectiveness is demonstrated in two settings: i) explicit likelihood modeling on binary MNIST, SVHN and CIFAR10, and ii) discrete latent modeling in an autoencoder trained on SVHN, CIFAR10 and Imagenet32. Empirically, we show considerable improvements over baselines in number of ARM inference calls and sampling speed.

In drug discovery, knowledge of the graph structure of chemical compounds is essential. Many thousands of scientific articles in chemistry and pharmaceutical sciences have investigated chemical compounds, but in cases the details of the structure of these chemical compounds is published only as an images. A tool to analyze these images automatically and convert them into a chemical graph structure would be useful for many applications, such drug discovery. A few such tools are available and they are mostly derived from optical character recognition. However, our evaluation of the performance of those tools reveals that they make often mistakes in detecting the correct bond multiplicity and stereochemical information. In addition, errors sometimes even lead to missing atoms in the resulting graph. In our work, we address these issues by developing a compound recognition method based on machine learning. More specifically, we develop a deep neural network model for optical compound recognition. The deep learning solution presented here consists of a segmentation model, followed by three classification models that predict atom locations, bonds and charges. Furthermore, this model not only predicts the graph structure of the molecule but also produces all information necessary to relate each component of the resulting graph to the source image. This solution is scalable and could rapidly process thousands of images. Finally, we compare empirically the proposed method to a well-established tool and observe significant error reductions.

The quality of a graph is determined jointly by three key factors of the graph: nodes, edges and similarity measure (or edge weights), and is very crucial to the success of graph-based semi-supervised learning (SSL) approaches. Recently, dynamic graph, which means part/all its factors are dynamically updated during the training process, has demonstrated to be promising for graph-based semi-supervised learning. However, existing approaches only update part of the three factors and keep the rest manually specified during the learning stage. In this paper, we propose a novel graph-based semi-supervised learning approach to optimize all three factors simultaneously in an end-to-end learning fashion. To this end, we concatenate two neural networks (feature network and similarity network) together to learn the categorical label and semantic similarity, respectively, and train the networks to minimize a unified SSL objective function. We also introduce an extended graph Laplacian regularization term to increase training efficiency. Extensive experiments on several benchmark datasets demonstrate the effectiveness of our approach.