JAX is a Python library designed for high-performance numerical computing, especially machine learning research. Its API for numerical functions is based on NumPy, a collection of functions used in scientific computing. Both Python and NumPy are widely used and familiar, making JAX simple, flexible, and easy to adopt. We have found that JAX has enabled rapid experimentation with novel algorithms and architectures and it now underpins many of our recent publications. To learn more please consider joining our JAX Roundtable, Wednesday December 9th 7:00pm GMT, at the NeurIPS virtual conference.

Advancements in artificial intelligence have paved the way for researchers and scientists to train machines to interpret images, sounds, and patterns in videos. Deep learning algorithms have given a new meaning to artistic explorations, which have stunned the world. AI is gradually advancing to become the next medium for art, and several artists worldwide relate to this seamless development. In a world that is evolving amid codes and hi-tech machines, Generative Adversarial Networks (GANs) are a specific class of algorithms being used to generate AI artworks. From pitch recognition to natural image synthesis and generating human faces that never existed-- machines are likely to match human senses.

With each passing year, models are getting more complex and bigger. A lot of AI models developed in research labs never see the light of the day. We can see that from 2012 to 2015, state-of-the-art Image recognition models went from being 8 layers to 152 layers. This begs the question is there a way to keep the model lighter and still able to process more complex data. The big challenge that we face with a bigger network is the training times.

In this work, we address the problem of unsupervised domain adaptation for person re-ID where annotations are available for the source domain but not for target. Previous methods typically follow a two-stage optimization pipeline, where the network is first pre-trained on source and then fine-tuned on target with pseudo labels created by feature clustering. Such methods sustain two main limitations. (1) The label noise may hinder the learning of discriminative features for recognizing target classes. (2) The domain gap may hinder knowledge transferring from source to target. We propose three types of technical schemes to alleviate these issues. First, we propose a cluster-wise contrastive learning algorithm (CCL) by iterative optimization of feature learning and cluster refinery to learn noise-tolerant representations in the unsupervised manner. Second, we adopt a progressive domain adaptation (PDA) strategy to gradually mitigate the domain gap between source and target data. Third, we propose Fourier augmentation (FA) for further maximizing the class separability of re-ID models by imposing extra constraints in the Fourier space. We observe that these proposed schemes are capable of facilitating the learning of discriminative feature representations. Experiments demonstrate that our method consistently achieves notable improvements over the state-of-the-art unsupervised re-ID methods on multiple benchmarks, e.g., surpassing MMT largely by 8.1\%, 9.9\%, 11.4\% and 11.1\% mAP on the Market-to-Duke, Duke-to-Market, Market-to-MSMT and Duke-to-MSMT tasks, respectively.

In contrast to image/text data whose order can be used to perform non-local feature aggregation in a straightforward way using the pooling layers, graphs lack the tensor representation and mostly the element-wise max/mean function is utilized to aggregate the locally extracted feature vectors. In this paper, we present a novel approach for global feature aggregation in Graph Neural Networks (GNNs) which utilizes a Latent Fixed Data Structure (LFDS) to aggregate the extracted feature vectors. The locally extracted feature vectors are sorted/distributed on the LFDS and a latent neural network (CNN/GNN) is utilized to perform feature aggregation on the LFDS. The proposed approach is used to design several novel global feature aggregation methods based on the choice of the LFDS. We introduce multiple LFDSs including loop, 3D tensor (image), sequence, data driven graphs and an algorithm which sorts/distributes the extracted local feature vectors on the LFDS. While the computational complexity of the proposed methods are linear with the order of input graphs, they achieve competitive or better results.

Perturbation-based attacks, while not physically realizable, have been the main emphasis of adversarial machine learning (ML) research. Patch-based attacks by contrast are physically realizable, yet most work has focused on 2D domain with recent forays into 3D. Characterizing the robustness properties of patch attacks and their invariance to 3D pose is important, yet not fully elucidated, and is the focus of this paper. To this end, several contributions are made here: A) we develop a new metric called mean Attack Success over Transformations (mAST) to evaluate patch attack robustness and invariance; and B), we systematically assess robustness of patch attacks to 3D position and orientation for various conditions; in particular, we conduct a sensitivity analysis which provides important qualitative insights into attack effectiveness as a function of the 3D pose of a patch relative to the camera (rotation, translation) and sets forth some properties for patch attack 3D invariance; and C), we draw novel qualitative conclusions including: 1) we demonstrate that for some 3D transformations, namely rotation and loom, increasing the training distribution support yields an increase in patch success over the full range at test time. 2) We provide new insights into the existence of a fundamental cutoff limit in patch attack effectiveness that depends on the extent of out-of-plane rotation angles. These findings should collectively guide future design of 3D patch attacks and defenses.

Data analysis has become a necessity in the modern era of cricket. Everything from effective team management to match win predictions use some form of analytics. Meaningful data representations are necessary for efficient analysis of data. In this study we investigate the use of adaptive (learnable) embeddings to represent inter-related features (such as players, teams, etc). The data used for this study is collected from a classical T20 tournament IPL (Indian Premier League). To naturally facilitate the learning of meaningful representations of features for accurate data analysis, we formulate a deep representation learning framework which jointly learns a custom set of embeddings (which represents our features of interest) through the minimization of a contrastive loss. We base our objective on a set of classes obtained as a result of hierarchical clustering on the overall run rate of an innings. It's been assessed that the framework ensures greater generality in the obtained embeddings, on top of which a task based analysis of overall run rate prediction was done to show the reliability of the framework.

Recently there have been many advances in research on language modeling of source code. Applications range from code suggestion and completion to code summarization. However, complete program synthesis of industry-grade programming languages has not been researched extensively. In this work, we introduce a variational autoencoder model for program synthesis of industry-grade programming languages. Our model incorporates the internal hierarchical structure of source codes and operates on parse trees. By learning a latent representation of source code over trees, we capture more information and achieve a higher performance than standard autoregressive autoencoder models. Furthermore, due to the tree-structured nature of our model, the autoregressive operations are performed on paths of trees instead of linear sequences. Therefore, the size of the sequences that the autoregressive model processes, scales proportionally to the width and depth of the tree instead of the total size of the tree which mitigates the common problem of exploding and vanishing gradients.

Semantic text matching is a critical problem in information retrieval. Recently, deep learning techniques have been widely used in this area and obtained significant performance improvements. However, most models are black boxes and it is hard to understand what happened in the matching process, due to the poor interpretability of deep learning. This paper aims at tackling this problem. The key idea is to test whether existing deep text matching methods satisfy some fundamental heuristics in information retrieval. Specifically, four heuristics are used in our study, i.e., term frequency constraint, term discrimination constraint, length normalization constraints, and TF-length constraint. Since deep matching models usually contain many parameters, it is difficult to conduct a theoretical study for these complicated functions. In this paper, We propose an empirical testing method. Specifically, We first construct some queries and documents to make them satisfy the assumption in a constraint, and then test to which extend a deep text matching model trained on the original dataset satisfies the corresponding constraint. Besides, a famous attribution based interpretation method, namely integrated gradient, is adopted to conduct detailed analysis and guide for feasible improvement. Experimental results on LETOR 4.0 and MS Marco show that all the investigated deep text matching methods, both representation and interaction based methods, satisfy the above constraints with high probabilities in statistics. We further extend these constraints to the semantic settings, which are shown to be better satisfied for all the deep text matching models. These empirical findings give clear understandings on why deep text matching models usually perform well in information retrieval. We believe the proposed evaluation methodology will be useful for testing future deep text matching models.

Spoken Language Understanding (SLU), a core component of the task-oriented dialogue system, expects a shorter inference latency due to the impatience of humans. Non-autoregressive SLU models clearly increase the inference speed but suffer uncoordinated-slot problems caused by the lack of sequential dependency information among each slot chunk. To gap this shortcoming, in this paper, we propose a novel non-autoregressive SLU model named Layered-Refine Transformer, which contains a Slot Label Generation (SLG) task and a Layered Refine Mechanism (LRM). SLG is defined as generating the next slot label with the token sequence and generated slot labels. With SLG, the non-autoregressive model can efficiently obtain dependency information during training and spend no extra time in inference. LRM predicts the preliminary SLU results from Transformer's middle states and utilizes them to guide the final prediction. Experiments on two public datasets indicate that our model significantly improves SLU performance (1.5\% on Overall accuracy) while substantially speed up (more than 10 times) the inference process over the state-of-the-art baseline.