AAAI 2021 论文列表

Abstract reasoning refers to the ability to analyze information, discover rules at an intangible level, and solve problems in innovative ways. Raven's Progressive Matrices (RPM) test is typically used to examine the capability of abstract reasoning. The subject is asked to identify the correct choice from the answer set to fill the missing panel at the bottom right of RPM (e.g., a 3×3 matrix), following the underlying rules inside the matrix. Recent studies, taking advantage of Convolutional Neural Networks (CNNs), have achieved encouraging progress to accomplish the RPM test. However, they partly ignore necessary inductive biases of RPM solver, such as order sensitivity within each row/column and incremental rule induction. To address this problem, in this paper we propose a Stratified Rule-Aware Network (SRAN) to generate the rule embeddings for two input sequences. Our SRAN learns multiple granularity rule embeddings at different levels, and incrementally integrates the stratified embedding flows through a gated fusion module. With the help of embeddings, a rule similarity metric is applied to guarantee that SRAN can not only be trained using a tuplet loss but also infer the best answer efficiently. We further point out the severe defects existing in the popular RAVEN dataset for RPM test, which prevent from the fair evaluation of the abstract reasoning ability. To fix the defects, we propose an answer set generation algorithm called Attribute Bisection Tree (ABT), forming an improved dataset named Impartial-RAVEN (I-RAVEN for short). Extensive experiments are conducted on both PGM and I-RAVEN datasets, showing that our SRAN outperforms the state-of-the-art models by a considerable margin.

We extend DatalogMTL—Datalog with operators from metric temporal logic—by adding stratified negation as failure. The new language provides additional expressive power for representing and reasoning about temporal data and knowledge in a wide range of applications. We consider models over the rational timeline, study their properties, and establish the computational complexity of reasoning. We show that, as in negation-free DatalogMTL, fact entailment in our language is PSPACE-complete in data and EXPSPACE-complete in combined complexity. Thus, the extension with stratified negation does not lead to higher complexity.

Spiking neural networks (SNNs) have great potential for energy-efficient implementation of Deep Neural Networks (DNNs) on dedicated neuromorphic hardware. Recent studies demonstrated competitive performance of SNNs compared with DNNs on image classification tasks, including CIFAR-10 and ImageNet data. The present work focuses on using SNNs in combination with deep reinforcement learning in ATARI games, which involves additional complexity as compared to image classification. We review the theory of converting DNNs to SNNs and extending the conversion to Deep Q-Networks (DQNs). We propose a robust representation of the firing rate to reduce the error during the conversion process. In addition, we introduce a new metric to evaluate the conversion process by comparing the decisions made by the DQN and SNN, respectively. We also analyze how the simulation time and parameter normalization influence the performance of converted SNNs. We achieve competitive scores on 17 top-performing Atari games. To the best of our knowledge, our work is the first to achieve state-of-the-art performance on multiple Atari games with SNNs. Our work serves as a benchmark for the conversion of DQNs to SNNs and paves the way for further research on solving reinforcement learning tasks with SNNs.

As an interesting and challenging task, story ending generation aims at generating a reasonable and coherent ending for a given story context. The key challenge of the task is to comprehend the context sufficiently and capture the hidden logic information effectively, which has not been well explored by most existing generative models. To tackle this issue, we propose a context-aware Multi-level Graph Convolutional Networks over Dependency Parse (MGCN-DP) trees to capture dependency relations and context clues more effectively. We utilize dependency parse trees to facilitate capturing relations and events in the context implicitly, and Multi-level Graph Convolutional Networks to update and deliver the representation crossing levels to obtain richer contextual information. Both automatic and manual evaluations show that our MGCN-DP can achieve comparable performance with state-of-the-art models. Our source code is available at https://github.com/VISLANG-Lab/MLGCN-DP.

Feature evolvable learning has been widely studied in recent years where old features will vanish and new features will emerge when learning with streams. Conventional methods usually assume that a label will be revealed after prediction at each time step. However, in practice, this assumption may not hold whereas no label will be given at most time steps. A good solution is to leverage the technique of manifold regularization to utilize the previous similar data to assist the refinement of the online model. Nevertheless, this approach needs to store all previous data which is impossible in learning with streams that arrive sequentially in large volume. Thus we need a buffer to store part of them. Considering that different devices may have different storage budgets, the learning approaches should be flexible subject to the storage budget limit. In this paper, we propose a new setting: Storage-Fit Feature-Evolvable streaming Learning (SF2EL) which incorporates the issue of rarely-provided labels into feature evolution. Our framework is able to fit its behavior for different storage budgets when learning with feature evolvable streams with unlabeled data. Besides, both theoretical and empirical results validate that our approach can preserve the merit of the original feature evolvable learning i.e., can always track the best baseline and thus perform well at any time step.

Quantitative trading and investment decision making are intricate financial tasks that rely on accurate stock selection. Despite advances in deep learning that have made significant progress in the complex and highly stochastic stock prediction problem, modern solutions face two significant limitations. They do not directly optimize the target of investment in terms of profit, and treat each stock as independent from the others, ignoring the rich signals between related stocks' temporal price movements. Building on these limitations, we reformulate stock prediction as a learning to rank problem and propose STHAN-SR, a neural hypergraph architecture for stock selection. The key novelty of our work is the proposal of modeling the complex relations between stocks through a hypergraph and a temporal Hawkes attention mechanism to tailor a new spatiotemporal attention hypergraph network architecture to rank stocks based on profit by jointly modeling stock interdependence and the temporal evolution of their prices. Through experiments on three markets spanning over six years of data, we show that STHAN-SR significantly outperforms state-of-the-art neural stock forecasting methods. We validate our design choices through ablative and exploratory analyses over STHAN-SR's spatial and temporal components and demonstrate its practical applicability.

The quantization of deep neural networks (QDNNs) has been actively studied for deployment in edge devices. Recent studies employ the knowledge distillation (KD) method to improve the performance of quantized networks. In this study, we propose stochastic precision ensemble training for QDNNs (SPEQ). SPEQ is a knowledge distillation training scheme; however, the teacher is formed by sharing the model parameters of the student network. We obtain the soft labels of the teacher by randomly changing the bit precision of the activation stochastically at each layer of the forward-pass computation. The student model is trained with these soft labels to reduce the activation quantization noise. The cosine similarity loss is employed, instead of the KL-divergence, for KD training. As the teacher model changes continuously by random bit-precision assignment, it exploits the effect of stochastic ensemble KD. SPEQ outperforms the existing quantization training methods in various tasks, such as image classification, question-answering, and transfer learning without the need for cumbersome teacher networks.

We study bandits with graph-structured feedback, where a learner repeatedly selects an arm and then observes rewards of the chosen arm as well as its neighbors in the feedback graph. Existing work on graphical bandits assumes either stochastic rewards or adversarial rewards, both of which are extremes and appear rarely in real-world scenarios. In this paper, we study graphical bandits with a reward model that interpolates between the two extremes, where the rewards are overall stochastically generated but a small fraction of them can be adversarially corrupted. For this problem, we propose an online algorithm that can utilize the stochastic pattern and also tolerate the adversarial corruptions. The main idea is to restrict exploration to carefully-designed independent sets of the feedback graph and perform exploitation by adopting a soft version of arm elimination. Theoretical analysis shows that our algorithm attains an $O(alpha ln{K} ln{T} + alpha C)$ regret, where $alpha$ is the independence number of the feedback graph, $K$ is the number of arms, $T$ is the time horizon, and $C$ quantifies the total corruptions introduced by the adversary. The effectiveness of our algorithm is demonstrated by numerical experiments.

We study a variant of stochastic bandits where the feedback model is specified by a graph. In this setting, after playing an arm, one can observe rewards of not only the played arm but also other arms that are adjacent to the played arm in the graph. Most of the existing work assumes the reward distributions are stationary over time, which, however, is often violated in common scenarios such as recommendation systems and online advertising. To address this limitation, we study stochastic bandits with graph feedback in non-stationary environments and propose algorithms with graph-dependent dynamic regret bounds. When the number of reward distribution changes L is known in advance, one of our algorithms achieves an Õ(√(αLT)) dynamic regret bound. We also develop an adaptive algorithm that can adapt to unknown L and attain an Õ(√(θLT)) dynamic regret. Here, α and θ are some graph-dependent quantities and T is the time horizon.

We study the effect of adversarial perturbations of images on the estimates of disparity by deep learning models trained for stereo. We show that imperceptible additive perturbations can significantly alter the disparity map, and correspondingly the perceived geometry of the scene. These perturbations not only affect the specific model they are crafted for, but transfer to models with different architecture, trained with different loss functions. We show that, when used for adversarial data augmentation, our perturbations result in trained models that are more robust, without sacrificing overall accuracy of the model. This is unlike what has been observed in image classification, where adding the perturbed images to the training set makes the model less vulnerable to adversarial perturbations, but to the detriment of overall accuracy. We test our method using the most recent stereo networks and evaluate their performance on public benchmark datasets.

Although the distributed machine learning methods can speed up the training of large deep neural networks, the communication cost has become the non-negligible bottleneck to constrain the performance. To address this challenge, the gradient compression based communication-efficient distributed learning methods were designed to reduce the communication cost, and more recently the local error feedback was incorporated to compensate for the corresponding performance loss. However, in this paper, we will show that a new "gradient mismatch" problem is raised by the local error feedback in centralized distributed training and can lead to degraded performance compared with full-precision training. To solve this critical problem, we propose two novel techniques, 1) step ahead and 2) error averaging, with rigorous theoretical analysis. Both our theoretical and empirical results show that our new methods can handle the "gradient mismatch" problem. The experimental results show that we can even train faster with common gradient compression schemes than both the full-precision training and local error feedback regarding the training epochs and without performance loss.

Forecasting influenza in a timely manner aids health organizations and policymakers in adequate preparation and decision making. However, effective influenza forecasting still remains a challenge despite increasing research interest. It is even more challenging amidst the COVID pandemic, when the influenza-like illness (ILI) counts are affected by various factors such as symptomatic similarities with COVID-19 and shift in healthcare seeking patterns of the general population. Under the current pandemic, historical influenza models carry valuable expertise about the disease dynamics but face difficulties adapting. Therefore, we propose CALI-Net, a neural transfer learning architecture which allows us to 'steer' a historical disease forecasting model to new scenarios where flu and COVID co-exist. Our framework enables this adaptation by automatically learning when it should emphasize learning from COVID-related signals and when it should learn from the historical model. Thus, we exploit representations learned from historical ILI data as well as the limited COVID-related signals. Our experiments demonstrate that our approach is successful in adapting a historical forecasting model to the current pandemic. In addition, we show that success in our primary goal, adaptation, does not sacrifice overall performance as compared with state-of-the-art influenza forecasting approaches.

Offline signature verification is a challenging issue that is widely used in various fields. Previous approaches model this task as a static feature matching or distance metric problem of two images. In this paper, we propose a novel Static-Dynamic Interaction Network (SDINet) model which introduces sequential representation into static signature images. A static signature image is converted to sequences by assuming pseudo dynamic processes in the static image. A static representation extracting deep features from signature images describes the global information of signatures. A dynamic representation extracting sequential features with LSTM networks characterizes the local information of signatures. A dynamic-to-static attention is learned from the sequences to refine the static features. Through the static-to-dynamic conversion and the dynamic-to-static attention, the static representation and dynamic representation are unified into a compact framework. The proposed method was evaluated on four popular datasets of different languages. The extensive experimental results manifest the strength of our model.

This paper focuses on a core task in computational sustainability and statistical ecology: species distribution modeling (SDM). In SDM, the occurrence pattern of a species on a landscape is predicted by environmental features based on observations at a set of locations. At first, SDM may appear to be a binary classification problem, and one might be inclined to employ classic tools (e.g., logistic regression, support vector machines, neural networks) to tackle it. However, wildlife surveys introduce structured noise (especially under-counting) in the species observations. If unaccounted for, these observation errors systematically bias SDMs. To address the unique challenges of SDM, this paper proposes a framework called StatEcoNet. Specifically, this work employs a graphical generative model in statistical ecology to serve as the skeleton of the proposed computational framework and carefully integrates neural networks under the framework. The advantages of StatEcoNet over related approaches are demonstrated on simulated datasets as well as bird species data. Since SDMs are critical tools for ecological science and natural resource management, StatEcoNet may offer boosted computational and analytical powers to a wide range of applications that have significant social impacts, e.g., the study and conservation of threatened species.

Few-shot detection and classification have advanced significantly in recent years. Yet, detection approaches require strong annotation (bounding boxes) both for pre-training and for adaptation to novel classes, and classification approaches rarely provide localization of objects in the scene. In this paper, we introduce StarNet - a few-shot model featuring an end-to-end differentiable non-parametric star-model detection and classification head. Through this head, the backbone is meta-trained using only image-level labels to produce good features for jointly localizing and classifying previously unseen categories of few-shot test tasks using a star-model that geometrically matches between the query and support images (to find corresponding object instances). Being a few-shot detector, StarNet does not require any bounding box annotations, neither during pre-training nor for novel classes adaptation. It can thus be applied to the previously unexplored and challenging task of Weakly Supervised Few-Shot Object Detection (WS-FSOD), where it attains significant improvements over the baselines. In addition, StarNet shows significant gains on few-shot classification benchmarks that are less cropped around the objects (where object localization is key).

Machine learning algorithms with empirical risk minimization are vulnerable under distributional shifts due to the greedy adoption of all the correlations found in training data. Recently, there are robust learning methods aiming at this problem by minimizing the worst-case risk over an uncertainty set. However, they equally treat all covariates to form the decision sets regardless of the stability of their correlations with the target, resulting in the overwhelmingly large set and low confidence of the learner. In this paper, we propose Stable Adversarial Learning (SAL) algorithm that leverages heterogeneous data sources to construct a more practical uncertainty set and conduct differentiated robustness optimization, where covariates are differentiated according to the stability of their correlations with the target. We theoretically show that our method is tractable for stochastic gradient-based optimization and provide the performance guarantees for our method. Empirical studies on both simulation and real datasets validate the effectiveness of our method in terms of uniformly good performance across unknown distributional shifts.

Learning complicated value functions in high dimensional state space by function approximation is a challenging task, partially due to that the max-operator used in temporal difference updates can theoretically cause instability for most linear or non-linear approximation schemes. Mellowmax is a recently proposed differentiable and non-expansion softmax operator that allows a convergent behavior in learning and planning. Unfortunately, the performance bound for the fixed point it converges to remains unclear, and in practice, its parameter is sensitive to various domains and has to be tuned case by case. Finally, the Mellowmax operator may suffer from oversmoothing as it ignores the probability being taken for each action when aggregating them. In this paper we address all the above issues with an enhanced Mellowmax operator, named SM2 (Soft Mellowmax). Particularly, the proposed operator is reliable, easy to implement, and has provable performance guarantee, while preserving all the advantages of Mellowmax. Furthermore, we show that our SM2 operator can be applied to the challenging multi-agent reinforcement learning scenarios, leading to stable value function approximation and state of the art performance.

The stability and generalization of stochastic gradient-based methods provide valuable insights into understanding the algorithmic performance of machine learning models. As the main workhorse for deep learning, the stochastic gradient descent has received a considerable amount of studies. Nevertheless, the community paid little attention to its decentralized variants. In this paper, we provide a novel formulation of the decentralized stochastic gradient descent. Leveraging this formulation together with (non)convex optimization theory, we establish the first stability and generalization guarantees for the decentralized stochastic gradient descent. Our theoretical results are built on top of a few common and mild assumptions and reveal that the decentralization deteriorates the stability of SGD for the first time. We verify our theoretical findings by using a variety of decentralized settings and benchmark machine learning models.

Continual learning has been a major problem in the deep learning community, where the main challenge is how to effectively learn a series of newly arriving tasks without forgetting the knowledge of previous tasks. Initiated by Learning without Forgetting (LwF), many of the existing works report that knowledge distillation is effective to preserve the previous knowledge, and hence they commonly use a soft label for the old task, namely a knowledge distillation (KD) loss, together with a class label for the new task, namely a cross entropy (CE) loss, to form a composite loss for a single neural network. However, this approach suffers from learning the knowledge by a CE loss as a KD loss often more strongly influences the objective function when they are in a competitive situation within a single network. This could be a critical problem particularly in a class incremental scenario, where the knowledge across tasks as well as within the new task, both of which can only be acquired by a CE loss, is essentially learned due to the existence of a unified classifier. In this paper, we propose a novel continual learning method, called Split-and-Bridge, which can successfully address the above problem by partially splitting a neural network into two partitions for training the new task separated from the old task and re-connecting them for learning the knowledge across tasks. In our thorough experimental analysis, our Split-and-Bridge method outperforms the state-of-the-art competitors in KD-based continual learning.