arXiv subjects

Qiang Liu

Publications and source records attributed to Qiang Liu.

At least 379 records · Page 21Linked to original sources

arXiv2021

Firefly Neural Architecture Descent: a General Approach for Growing Neural Networks

We propose firefly neural architecture descent, a general framework for progressively and dynamically growing neural networks to jointly optimize the networks' parameters and architectures. Our method works in a steepest descent fashion, which iteratively finds the best network within a functional neighborhood of the original network that includes a diverse set of candidate network structures. By using Taylor approximation, the optimal network structure in the neighborhood can be found with a greedy selection procedure. We show that firefly descent can flexibly grow networks both wider and deeper, and can be applied to learn accurate but resource-efficient neural architectures that avoid catastrophic forgetting in continual learning. Empirically, firefly descent achieves promising results on both neural architecture search and continual learning. In particular, on a challenging continual image classification task, it learns networks that are smaller in size but have higher average accuracy than those learned by the state-of-the-art methods.

cs.LG↗

arXiv2021

Vision Transformers with Patch Diversification

Vision transformer has demonstrated promising performance on challenging computer vision tasks. However, directly training the vision transformers may yield unstable and sub-optimal results. Recent works propose to improve the performance of the vision transformers by modifying the transformer structures, e.g., incorporating convolution layers. In contrast, we investigate an orthogonal approach to stabilize the vision transformer training without modifying the networks. We observe the instability of the training can be attributed to the significant similarity across the extracted patch representations. More specifically, for deep vision transformers, the self-attention blocks tend to map different patches into similar latent representations, yielding information loss and performance degradation. To alleviate this problem, in this work, we introduce novel loss functions in vision transformer training to explicitly encourage diversity across patch representations for more discriminative feature extraction. We empirically show that our proposed techniques stabilize the training and allow us to train wider and deeper vision transformers. We further show the diversified features significantly benefit the downstream tasks in transfer learning. For semantic segmentation, we enhance the state-of-the-art (SOTA) results on Cityscapes and ADE20k. Our code is available at https://github.com/ChengyueGongR/PatchVisionTransformer.

cs.CV↗

arXiv2021

AlphaNet: Improved Training of Supernets with Alpha-Divergence

Weight-sharing neural architecture search (NAS) is an effective technique for automating efficient neural architecture design. Weight-sharing NAS builds a supernet that assembles all the architectures as its sub-networks and jointly trains the supernet with the sub-networks. The success of weight-sharing NAS heavily relies on distilling the knowledge of the supernet to the sub-networks. However, we find that the widely used distillation divergence, i.e., KL divergence, may lead to student sub-networks that over-estimate or under-estimate the uncertainty of the teacher supernet, leading to inferior performance of the sub-networks. In this work, we propose to improve the supernet training with a more generalized alpha-divergence. By adaptively selecting the alpha-divergence, we simultaneously prevent the over-estimation or under-estimation of the uncertainty of the teacher model. We apply the proposed alpha-divergence based supernets training to both slimmable neural networks and weight-sharing NAS, and demonstrate significant improvements. Specifically, our discovered model family, AlphaNet, outperforms prior-art models on a wide range of FLOPs regimes, including BigNAS, Once-for-All networks, and AttentiveNAS. We achieve ImageNet top-1 accuracy of 80.0% with only 444M FLOPs. Our code and pretrained models are available at https://github.com/facebookresearch/AlphaNet.

cs.CV↗

arXiv2021

Any equation is a forest: Symbolic genetic algorithm for discovering open-form partial differential equations (SGA-PDE)

Partial differential equations (PDEs) are concise and understandable representations of domain knowledge, which are essential for deepening our understanding of physical processes and predicting future responses. However, the PDEs of many real-world problems are uncertain, which calls for PDE discovery. We propose the symbolic genetic algorithm (SGA-PDE) to discover open-form PDEs directly from data without prior knowledge about the equation structure. SGA-PDE focuses on the representation and optimization of PDE. Firstly, SGA-PDE uses symbolic mathematics to realize the flexible representation of any given PDE, transforms a PDE into a forest, and converts each function term into a binary tree. Secondly, SGA-PDE adopts a specially designed genetic algorithm to efficiently optimize the binary trees by iteratively updating the tree topology and node attributes. The SGA-PDE is gradient-free, which is a desirable characteristic in PDE discovery since it is difficult to obtain the gradient between the PDE loss and the PDE structure. In the experiment, SGA-PDE not only successfully discovered nonlinear Burgers' equation, Korteweg-de Vries (KdV) equation, and Chafee-Infante equation, but also handled PDEs with fractional structure and compound functions that cannot be solved by conventional PDE discovery methods.

cs.NE↗

arXiv2021

Point Spread Function Estimation for Wide Field Small Aperture Telescopes with Deep Neural Networks and Calibration Data

The point spread function (PSF) reflects states of a telescope and plays an important role in development of data processing methods, such as PSF based astrometry, photometry and image restoration. However, for wide field small aperture telescopes (WFSATs), estimating PSF in any position of the whole field of view is hard, because aberrations induced by the optical system are quite complex and the signal to noise ratio of star images is often too low for PSF estimation. In this paper, we further develop our deep neural network (DNN) based PSF modelling method and show its applications in PSF estimation. During the telescope alignment and testing stage, our method collects system calibration data through modification of optical elements within engineering tolerances (tilting and decentering). Then we use these data to train a DNN (Tel--Net). After training, the Tel--Net can estimate PSF in any field of view from several discretely sampled star images. We use both simulated and experimental data to test performance of our method. The results show that the Tel--Net can successfully reconstruct PSFs of WFSATs of any states and in any positions of the FoV. Its results are significantly more precise than results obtained by the compared classic method - Inverse Distance Weight (IDW) interpolation. Our method provides foundations for developing of deep neural network based data processing methods for WFSATs, which require strong prior information of PSFs.

astro-ph.IM↗

arXiv2021

Eudoxus: Characterizing and Accelerating Localization in Autonomous Machines

We develop and commercialize autonomous machines, such as logistic robots and self-driving cars, around the globe. A critical challenge to our -- and any -- autonomous machine is accurate and efficient localization under resource constraints, which has fueled specialized localization accelerators recently. Prior acceleration efforts are point solutions in that they each specialize for a specific localization algorithm. In real-world commercial deployments, however, autonomous machines routinely operate under different environments and no single localization algorithm fits all the environments. Simply stacking together point solutions not only leads to cost and power budget overrun, but also results in an overly complicated software stack. This paper demonstrates our new software-hardware co-designed framework for autonomous machine localization, which adapts to different operating scenarios by fusing fundamental algorithmic primitives. Through characterizing the software framework, we identify ideal acceleration candidates that contribute significantly to the end-to-end latency and/or latency variation. We show how to co-design a hardware accelerator to systematically exploit the parallelisms, locality, and common building blocks inherent in the localization framework. We build, deploy, and evaluate an FPGA prototype on our next-generation self-driving cars. To demonstrate the flexibility of our framework, we also instantiate another FPGA prototype targeting drones, which represent mobile autonomous machines. We achieve about 2x speedup and 4x energy reduction compared to widely-deployed, optimized implementations on general-purpose platforms.

cs.AR↗

arXiv2021

When Contrastive Learning Meets Active Learning: A Novel Graph Active Learning Paradigm with Self-Supervision

This paper studies active learning (AL) on graphs, whose purpose is to discover the most informative nodes to maximize the performance of graph neural networks (GNNs). Previously, most graph AL methods focus on learning node representations from a carefully selected labeled dataset with large amount of unlabeled data neglected. Motivated by the success of contrastive learning (CL), we propose a novel paradigm that seamlessly integrates graph AL with CL. While being able to leverage the power of abundant unlabeled data in a self-supervised manner, nodes selected by AL further provide semantic information that can better guide representation learning. Besides, previous work measures the informativeness of nodes without considering the neighborhood propagation scheme of GNNs, so that noisy nodes may be selected. We argue that due to the smoothing nature of GNNs, the central nodes from homophilous subgraphs should benefit the model training most. To this end, we present a minimax selection scheme that explicitly harnesses neighborhood information and discover homophilous subgraphs to facilitate active selection. Comprehensive, confounding-free experiments on five public datasets demonstrate the superiority of our method over state-of-the-arts.

cs.LG↗

arXiv2021

DyGCN: Dynamic Graph Embedding with Graph Convolutional Network

Graph embedding, aiming to learn low-dimensional representations (aka. embeddings) of nodes, has received significant attention recently. Recent years have witnessed a surge of efforts made on static graphs, among which Graph Convolutional Network (GCN) has emerged as an effective class of models. However, these methods mainly focus on the static graph embedding. In this work, we propose an efficient dynamic graph embedding approach, Dynamic Graph Convolutional Network (DyGCN), which is an extension of GCN-based methods. We naturally generalizes the embedding propagation scheme of GCN to dynamic setting in an efficient manner, which is to propagate the change along the graph to update node embeddings. The most affected nodes are first updated, and then their changes are propagated to the further nodes and leads to their update. Extensive experiments conducted on various dynamic graphs demonstrate that our model can update the node embeddings in a time-saving and performance-preserving way.

cs.LG↗

arXiv2021

Disentangled Item Representation for Recommender Systems

Item representations in recommendation systems are expected to reveal the properties of items. Collaborative recommender methods usually represent an item as one single latent vector. Nowadays the e-commercial platforms provide various kinds of attribute information for items (e.g., category, price and style of clothing). Utilizing these attribute information for better item representations is popular in recent years. Some studies use the given attribute information as side information, which is concatenated with the item latent vector to augment representations. However, the mixed item representations fail to fully exploit the rich attribute information or provide explanation in recommender systems. To this end, we propose a fine-grained Disentangled Item Representation (DIR) for recommender systems in this paper, where the items are represented as several separated attribute vectors instead of a single latent vector. In this way, the items are represented at the attribute level, which can provide fine-grained information of items in recommendation. We introduce a learning strategy, LearnDIR, which can allocate the corresponding attribute vectors to items. We show how DIR can be applied to two typical models, Matrix Factorization (MF) and Recurrent Neural Network (RNN). Experimental results on two real-world datasets show that the models developed under the framework of DIR are effective and efficient. Even using fewer parameters, the proposed model can outperform the state-of-the-art methods, especially in the cold-start situation. In addition, we make visualizations to show that our proposition can provide explanation for users in real-world applications.

cs.IR↗

arXiv2021

VCNet and Functional Targeted Regularization For Learning Causal Effects of Continuous Treatments

Motivated by the rising abundance of observational data with continuous treatments, we investigate the problem of estimating the average dose-response curve (ADRF). Available parametric methods are limited in their model space, and previous attempts in leveraging neural network to enhance model expressiveness relied on partitioning continuous treatment into blocks and using separate heads for each block; this however produces in practice discontinuous ADRFs. Therefore, the question of how to adapt the structure and training of neural network to estimate ADRFs remains open. This paper makes two important contributions. First, we propose a novel varying coefficient neural network (VCNet) that improves model expressiveness while preserving continuity of the estimated ADRF. Second, to improve finite sample performance, we generalize targeted regularization to obtain a doubly robust estimator of the whole ADRF curve.

cs.LG↗

arXiv2021

Non-asymptotic Confidence Intervals of Off-policy Evaluation: Primal and Dual Bounds

Off-policy evaluation (OPE) is the task of estimating the expected reward of a given policy based on offline data previously collected under different policies. Therefore, OPE is a key step in applying reinforcement learning to real-world domains such as medical treatment, where interactive data collection is expensive or even unsafe. As the observed data tends to be noisy and limited, it is essential to provide rigorous uncertainty quantification, not just a point estimation, when applying OPE to make high stakes decisions. This work considers the problem of constructing non-asymptotic confidence intervals in infinite-horizon off-policy evaluation, which remains a challenging open question. We develop a practical algorithm through a primal-dual optimization-based approach, which leverages the kernel Bellman loss (KBL) of Feng et al.(2019) and a new martingale concentration inequality of KBL applicable to time-dependent data with unknown mixing conditions. Our algorithm makes minimum assumptions on the data and the function class of the Q-function, and works for the behavior-agnostic settings where the data is collected under a mix of arbitrary unknown behavior policies. We present empirical results that clearly demonstrate the advantages of our approach over existing methods.

cs.LG↗

arXiv2021

Centroid Transformers: Learning to Abstract with Attention

Self-attention, as the key block of transformers, is a powerful mechanism for extracting features from the inputs. In essence, what self-attention does is to infer the pairwise relations between the elements of the inputs, and modify the inputs by propagating information between input pairs. As a result, it maps inputs to N outputs and casts a quadratic $O(N^2)$ memory and time complexity. We propose centroid attention, a generalization of self-attention that maps N inputs to M outputs $(M\leq N)$, such that the key information in the inputs are summarized in the smaller number of outputs (called centroids). We design centroid attention by amortizing the gradient descent update rule of a clustering objective function on the inputs, which reveals an underlying connection between attention and clustering. By compressing the inputs to the centroids, we extract the key information useful for prediction and also reduce the computation of the attention module and the subsequent layers. We apply our method to various applications, including abstractive text summarization, 3D vision, and image processing. Empirical results demonstrate the effectiveness of our method over the standard transformers.

cs.LG↗

arXiv2021

Graph Contrastive Learning with Adaptive Augmentation

Recently, contrastive learning (CL) has emerged as a successful method for unsupervised graph representation learning. Most graph CL methods first perform stochastic augmentation on the input graph to obtain two graph views and maximize the agreement of representations in the two views. Despite the prosperous development of graph CL methods, the design of graph augmentation schemes -- a crucial component in CL -- remains rarely explored. We argue that the data augmentation schemes should preserve intrinsic structures and attributes of graphs, which will force the model to learn representations that are insensitive to perturbation on unimportant nodes and edges. However, most existing methods adopt uniform data augmentation schemes, like uniformly dropping edges and uniformly shuffling features, leading to suboptimal performance. In this paper, we propose a novel graph contrastive representation learning method with adaptive augmentation that incorporates various priors for topological and semantic aspects of the graph. Specifically, on the topology level, we design augmentation schemes based on node centrality measures to highlight important connective structures. On the node attribute level, we corrupt node features by adding more noise to unimportant node features, to enforce the model to recognize underlying semantic information. We perform extensive experiments of node classification on a variety of real-world datasets. Experimental results demonstrate that our proposed method consistently outperforms existing state-of-the-art baselines and even surpasses some supervised counterparts, which validates the effectiveness of the proposed contrastive framework with adaptive augmentation.

cs.LG↗

arXiv2021

DNN2LR: Automatic Feature Crossing for Credit Scoring

Credit scoring is a major application of machine learning for financial institutions to decide whether to approve or reject a credit loan. For sake of reliability, it is necessary for credit scoring models to be both accurate and globally interpretable. Simple classifiers, e.g., Logistic Regression (LR), are white-box models, but not powerful enough to model complex nonlinear interactions among features. Fortunately, automatic feature crossing is a promising way to find cross features to make simple classifiers to be more accurate without heavy handcrafted feature engineering. However, credit scoring is usually based on different aspects of users, and the data usually contains hundreds of feature fields. This makes existing automatic feature crossing methods not efficient for credit scoring. In this work, we find local piece-wise interpretations in Deep Neural Networks (DNNs) of a specific feature are usually inconsistent in different samples, which is caused by feature interactions in the hidden layers. Accordingly, we can design an automatic feature crossing method to find feature interactions in DNN, and use them as cross features in LR. We give definition of the interpretation inconsistency in DNN, based on which a novel feature crossing method for credit scoring prediction called DNN2LR is proposed. Apparently, the final model, i.e., a LR model empowered with cross features, generated by DNN2LR is a white-box model. Extensive experiments have been conducted on both public and business datasets from real-world credit scoring applications. Experimental shows that, DNN2LR can outperform the DNN model, as well as several feature crossing methods. Moreover, comparing with the state-of-the-art feature crossing methods, i.e., AutoCross, DNN2LR can accelerate the speed for feature crossing by about 10 to 40 times on datasets with large numbers of feature fields.

cs.LG↗

arXiv2021

STAN: Spatio-Temporal Attention Network for Next Location Recommendation

The next location recommendation is at the core of various location-based applications. Current state-of-the-art models have attempted to solve spatial sparsity with hierarchical gridding and model temporal relation with explicit time intervals, while some vital questions remain unsolved. Non-adjacent locations and non-consecutive visits provide non-trivial correlations for understanding a user's behavior but were rarely considered. To aggregate all relevant visits from user trajectory and recall the most plausible candidates from weighted representations, here we propose a Spatio-Temporal Attention Network (STAN) for location recommendation. STAN explicitly exploits relative spatiotemporal information of all the check-ins with self-attention layers along the trajectory. This improvement allows a point-to-point interaction between non-adjacent locations and non-consecutive check-ins with explicit spatiotemporal effect. STAN uses a bi-layer attention architecture that firstly aggregates spatiotemporal correlation within user trajectory and then recalls the target with consideration of personalized item frequency (PIF). By visualization, we show that STAN is in line with the above intuition. Experimental results unequivocally show that our model outperforms the existing state-of-the-art methods by 9-17%.

cs.IR↗

arXiv2021

Coherent ray-wave structured light based on (helical) Ince-Gaussian modes

The topological evolution of classic eigenmodes including Hermite-Laguerre-Gaussian and (helical) InceGaussian modes is exploited to construct coherent state modes, which unifies the representations of travelingwave (TW) and standing-wave (SW) ray-wave structured light for the first time and realizes the TW-SW unified ray-wave geometric beam with topology of raytrajectories splitting effect, breaking the boundary of TW and SW structured light. We experimentally generate these new modes with high purity and dynamic control by digital holography method, revealing potential applications in optical manipulation and communication.

physics.optics↗

arXiv2021

DNN2LR: Interpretation-inspired Feature Crossing for Real-world Tabular Data

For sake of reliability, it is necessary for models in real-world applications to be both powerful and globally interpretable. Simple classifiers, e.g., Logistic Regression (LR), are globally interpretable, but not powerful enough to model complex nonlinear interactions among features in tabular data. Meanwhile, Deep Neural Networks (DNNs) have shown great effectiveness for modeling tabular data, but is not globally interpretable. In this work, we find local piece-wise interpretations in DNN of a specific feature are usually inconsistent in different samples, which is caused by feature interactions in the hidden layers. Accordingly, we can design an automatic feature crossing method to find feature interactions in DNN, and use them as cross features in LR. We give definition of the interpretation inconsistency in DNN, based on which a novel feature crossing method called DNN2LR is proposed. Extensive experiments have been conducted on four public datasets and two real-world datasets. The final model, i.e., a LR model empowered with cross features, generated by DNN2LR can outperform the complex DNN model, as well as several state-of-the-art feature crossing methods. The experimental results strongly verify the effectiveness and efficiency of DNN2LR, especially on real-world datasets with large numbers of feature fields.

cs.LG↗

arXiv2021

Post-training Quantization with Multiple Points: Mixed Precision without Mixed Precision

We consider the post-training quantization problem, which discretizes the weights of pre-trained deep neural networks without re-training the model. We propose multipoint quantization, a quantization method that approximates a full-precision weight vector using a linear combination of multiple vectors of low-bit numbers; this is in contrast to typical quantization methods that approximate each weight using a single low precision number. Computationally, we construct the multipoint quantization with an efficient greedy selection procedure, and adaptively decides the number of low precision points on each quantized weight vector based on the error of its output. This allows us to achieve higher precision levels for important weights that greatly influence the outputs, yielding an 'effect of mixed precision' but without physical mixed precision implementations (which requires specialized hardware accelerators). Empirically, our method can be implemented by common operands, bringing almost no memory and computation overhead. We show that our method outperforms a range of state-of-the-art methods on ImageNet classification and it can be generalized to more challenging tasks like PASCAL VOC object detection.

cs.LG↗