Large natural language models (such as GPT-3 or T5) demonstrate impressive abilities across a range of general NLP tasks. Here, we show that the knowledge embedded in such models provides a useful inductive bias, not just on traditional NLP tasks, but also in the nontraditional task of training a symbolic reasoning engine. We observe that these engines learn quickly and generalize in a natural way that reflects human intuition. For example, training such a system to model block-stacking might naturally generalize to stacking other types of objects because of structure in the real world that has been partially captured by the language describing it. We study several abstract textual reasoning tasks, such as object manipulation and navigation, and demonstrate multiple types of generalization to novel scenarios and the symbols that comprise them. We also demonstrate the surprising utility of \textit{compositional learning}, where a learner dedicated to mastering a complicated task gains an advantage by training on relevant simpler tasks instead of jumping straight to the complicated task.

A recent line of work [3, 5, 23, 29] has established intriguing connections between the generalization/compression properties of a deep neural network (DNN) model and the so-called layer weights' stable ranks. Intuitively, the latter are indicators of the effective number of parameters in the net. In this work, we address some natural questions regarding the space of DNNs conditioned on the layers' stable rank, where we study feed-forward dynamics, initialization, training and expressivity. To this end, we first propose a random DNN model with a new sampling scheme based on stable rank. Then, we show how feed-forward maps are affected by the constraint and how training evolves in the overparametrized regime (via Neural Tangent Kernels). Our results imply that stable ranks appear layerwise essentially as linear factors whose effect accumulates exponentially depthwise. Moreover, we provide empirical analysis suggesting that stable rank initialization alone can lead to convergence speed ups.

A program written with differentiable programming can be differentiated automatically. The differentiation results can then be used for gradient-based optimization (e.g., gradient descent) of the parameters in the program. Differentiable programming have been used in scientific computing, physics simulations, and other domains to help mitigate the burden of manual error-prone coding of derivative computations. Recent several years have witnessed a growing interest of differentiable programming in machine learning (ML) [11, 34] and Probabilistic Programming [30], to accommodate the needs of various customized ML operators, user-defined operations in the learning targets (e.g., the physical environment of reinforcement learning) and statistical sampling. The key technique in differentiable programming is automatic differentiation. For a program (P) that produces output (y) from some given values (X), automatic differentiation automatically computes the derivatives ( y/ x) (x X) without the need for users to write the differentiation code. The given program P is called the primal code, and x is called an active input variable. Existing approaches of automatic differentiation fall into two categories: (i) Symbolic differentiation, which uses expression manipulation in computer algebra systems, (ii) Algorithmic differentiation, which performs a non-standard interpretation of a given computer program by replacing the domain of the variables to incorporate derivative values and redefining the semantics of the operators to propagate derivatives per the chain rule of differential calculus (elaborated in Section 2). Symbolic differentiation has been commonly regarded inappropriate for differentiable programming, for several reasons: (i) It results in complex and cryptic expressions plagued with the problem of "expression swell" [5].

We introduce a method for unsupervised parsing that relies on bootstrapping classifiers to identify if a node dominates a specific span in a sentence. There are two types of classifiers, an inside classifier that acts on a span, and an outside classifier that acts on everything outside of a given span. Through self-training and co-training with the two classifiers, we show that the interplay between them helps improve the accuracy of both, and as a result, effectively parse. A seed bootstrapping technique prepares the data to train these classifiers. Our analyses further validate that such an approach in conjunction with weak supervision using prior branching knowledge of a known language (left/right-branching) and minimal heuristics injects strong inductive bias into the parser, achieving 63.1 F$_1$ on the English (PTB) test set. In addition, we show the effectiveness of our architecture by evaluating on treebanks for Chinese (CTB) and Japanese (KTB) and achieve new state-of-the-art results.\footnote{For code or data, please contact the authors.}

Convolutions are fundamental elements in deep learning architectures. Here, we present a theoretical framework for combining extrinsic and intrinsic approaches to manifold convolution through isometric embeddings into tori. In this way, we define a convolution operator for a manifold of arbitrary topology and dimension. We also explain geometric and topological conditions that make some local definitions of convolutions which rely on translating filters along geodesic paths on a manifold, computationally intractable. A result of Alan Turing from 1938 underscores the need for such a toric isometric embedding approach to achieve a global definition of convolution on computable, finite metric space approximations to a smooth manifold.

Region proposal based methods like R-CNN and Faster R-CNN models have proven to be extremely successful in object detection and segmentation tasks. Recently, Transformers have also gained popularity in the domain of Computer Vision, and are being utilised to improve the performance of conventional models. In this paper, we present a relatively new transformer based approach to enhance the performance of the conventional convolutional feature extractor in the existing region proposal based methods. Our approach merges the convolutional feature maps with transformer-based token embeddings by applying a projection operation similar to self-attention in transformers. The results of our experiments show that transformer assisted feature extractor achieves a significant improvement in mIoU (mean Intersection over Union) scores compared to vanilla convolutional backbone.

Light-weight convolutional neural networks (CNNs) are the de-facto for mobile vision tasks. Their spatial inductive biases allow them to learn representations with fewer parameters across different vision tasks. However, these networks are spatially local. To learn global representations, self-attention-based vision transformers (ViTs) have been adopted. Unlike CNNs, ViTs are heavy-weight. In this paper, we ask the following question: is it possible to combine the strengths of CNNs and ViTs to build a light-weight and low latency network for mobile vision tasks? Towards this end, we introduce MobileViT, a light-weight and generalpurpose vision transformer for mobile devices. MobileViT presents a different perspective for the global processing of information with transformers, i.e., transformers as convolutions. Our results show that MobileViT significantly outperforms CNN-and ViT-based networks across different tasks and datasets. On the ImageNet-1k dataset, MobileViT achieves top-1 accuracy of 78.4% with about 6 million parameters, which is 3.2% and 6.2% more accurate than MobileNetv3 (CNN-based) and DeIT (ViT-based) for a similar number of parameters. On the MS-COCO object detection task, MobileViT is 5.7% more accurate than MobileNetv3 for a similar number of parameters. Self-attention-based models, especially vision transformers (ViTs; Figure 1a; Dosovitskiy et al., 2021), are an alternative to convolutional neural networks (CNNs) to learn visual representations.

Distributed stochastic gradient descent (SGD) approach has been widely used in large-scale deep learning, and the gradient collective method is vital to ensure the training scalability of the distributed deep learning system. Collective communication such as AllReduce has been widely adopted for the distributed SGD process to reduce the communication time. However, AllReduce incurs large bandwidth resources while most gradients are sparse in many cases since many gradient values are zeros and should be efficiently compressed for bandwidth saving. To reduce the sparse gradient communication overhead, we propose Sparse-Sketch Reducer (S2 Reducer), a novel sketch-based sparse gradient aggregation method with convergence guarantees. S2 Reducer reduces the communication cost by only compressing the non-zero gradients with count-sketch and bitmap, and enables the efficient AllReduce operators for parallel SGD training. We perform extensive evaluation against four state-of-the-art methods over five training models. Our results show that S2 Reducer converges to the same accuracy, reduces 81\% sparse communication overhead, and achieves 1.8$ \times $ speedup compared to state-of-the-art approaches.

Text classification is a primary task in natural language processing (NLP). Recently, graph neural networks (GNNs) have developed rapidly and been applied to text classification tasks. Although more complex models tend to achieve better performance, research highly depends on the computing power of the device used. In this article, we propose TENT (https://github.com/Daisean/TENT) to obtain better text classification performance and reduce the reliance on computing power. Specifically, we first establish a dependency analysis graph for each text and then convert each graph into its corresponding encoding tree. The representation of the entire graph is obtained by updating the representation of the non-leaf nodes in the encoding tree. Experimental results show that our method outperforms other baselines on several datasets while having a simple structure and few parameters.

Coronavirus disease (COVID-19) is an infectious respiratory disease that was first discovered in late December 2019, in Wuhan, China, and then spread worldwide causing a lot of panic and death. Users of social networking sites such as Facebook and Twitter have been focused on reading, publishing, and sharing novelties, tweets, and articles regarding the newly emerging pandemic. A lot of these users often employ sarcasm to convey their intended meaning in a humorous, funny, and indirect way making it hard for computer-based applications to automatically understand and identify their goal and the harm level that they can inflect. Motivated by the emerging need for annotated datasets that tackle these kinds of problems in the context of COVID-19, this paper builds and releases AraCOVID19-SSD a manually annotated Arabic COVID-19 sarcasm and sentiment detection dataset containing 5,162 tweets. To confirm the practical utility of the built dataset, it has been carefully analyzed and tested using several classification models.