AAAI 2023 Papers

Pretrained code language models have enabled great progress towards program synthesis. However, common approaches only consider in-file local context and thus miss information and constraints imposed by other parts of the codebase and its external dependencies. Existing code completion benchmarks also lack such context. To resolve these restrictions we curate a new dataset of permissively licensed Python packages that includes full projects and their dependencies and provide tools to extract non-local information with the help of program analyzers. We then focus on the task of function call argument completion which requires predicting the arguments to function calls. We show that existing code completion models do not yield good results on our completion task. To better solve this task, we query a program analyzer for information relevant to a given function call, and consider ways to provide the analyzer results to different code completion models during inference and training. Our experiments show that providing access to the function implementation and function usages greatly improves the argument completion performance. Our ablation study provides further insights on how different types of information available from the program analyzer and different ways of incorporating the information affect the model performance.

As deep learning models become popular, there is a lot of need for deploying them to diverse device environments. Because it is costly to develop and optimize a neural network for every single environment, there is a line of research to search neural networks for multiple target environments efficiently. However, existing works for such a situation still suffer from requiring many GPUs and expensive costs. Motivated by this, we propose a novel neural network optimization framework named Bespoke for low-cost deployment. Our framework searches for a lightweight model by replacing parts of an original model with randomly selected alternatives, each of which comes from a pretrained neural network or the original model. In the practical sense, Bespoke has two significant merits. One is that it requires near zero cost for designing the search space of neural networks. The other merit is that it exploits the sub-networks of public pretrained neural networks, so the total cost is minimal compared to the existing works. We conduct experiments exploring Bespoke's the merits, and the results show that it finds efficient models for multiple targets with meager cost.

We consider a sequential decision making problem where the agent faces the environment characterized by the stochastic discrete events and seeks an optimal intervention policy such that its long-term reward is maximized. This problem exists ubiquitously in social media, finance and health informatics but is rarely investigated by the conventional research in reinforcement learning. To this end, we present a novel framework of the model-based reinforcement learning where the agent's actions and observations are asynchronous stochastic discrete events occurring in continuous-time. We model the dynamics of the environment by Hawkes process with external intervention control term and develop an algorithm to embed such process in the Bellman equation which guides the direction of the value gradient. We demonstrate the superiority of our method in both synthetic simulator and real-data experiments.

This paper studies the problem of multi-step manipulative attacks in Stackelberg security games, in which a clever attacker attempts to orchestrate its attacks over multiple time steps to mislead the defender's learning of the attacker's behavior. This attack manipulation eventually influences the defender's patrol strategy towards the attacker's benefit. Previous work along this line of research only focuses on one-shot games in which the defender learns the attacker's behavior and then designs a corresponding strategy only once. Our work, on the other hand, investigates the long-term impact of the attacker's manipulation in which current attack and defense choices of players determine the future learning and patrol planning of the defender. This paper has three key contributions. First, we introduce a new multi-step manipulative attack game model that captures the impact of sequential manipulative attacks carried out by the attacker over the entire time horizon. Second, we propose a new algorithm to compute an optimal manipulative attack plan for the attacker, which tackles the challenge of multiple connected optimization components involved in the computation across multiple time steps. Finally, we present extensive experimental results on the impact of such misleading attacks, showing a significant benefit for the attacker and loss for the defender.

Offline reinforcement learning (RL) have received rising interest due to its appealing data efficiency. The present study addresses behavior estimation, a task that aims at estimating the data-generating policy. In particular, this work considers a scenario where data are collected from multiple sources. Neglecting data heterogeneity, existing approaches cannot provide good estimates and impede policy learning. To overcome this drawback, the present study proposes a latent variable model and a model-learning algorithm to infer a set of policies from data, which allows an agent to use as behavior policy the policy that best describes a particular trajectory. To illustrate the benefit of such a fine-grained characterization for multi-source data, this work showcases how the proposed model can be incorporated into an existing offline RL algorithm. Lastly, with extensive empirical evaluation this work confirms the risks of neglecting data heterogeneity and the efficacy of the proposed model.

Active Learning is an essential method for label-efficient deep learning. As a Bayesian active learning method, Bayesian Active Learning by Disagreement (BALD) successfully selects the most representative samples by maximizing the mutual information between the model prediction and model parameters. However, when applied to a batch acquisition mode, like batch construction with greedy search, BALD suffers from poor performance, especially with noises of near-duplicate data. To address this shortcoming, we propose a diverse beam search optimized batch active learning method, which explores a graph for every batch construction by expanding the highest-scored samples of a predetermined number. To avoid near duplicate beam branches (very similar beams generated from the same root and similar samples), which is undesirable for lacking diverse representations in the feature space, we design a self-adapted constraint within candidate beams. The proposed method is able to acquire data that can better represent the distribution of the unlabeled pool, and at the same time, be significantly different from existing beams. We observe that the proposed method achieves higher batch performance than the baseline methods on three benchmark datasets.

We study the combinatorial assignment domain, which includes combinatorial auctions and course allocation. The main challenge in this domain is that the bundle space grows exponentially in the number of items. To address this, several papers have recently proposed machine learning-based preference elicitation algorithms that aim to elicit only the most important information from agents. However, the main shortcoming of this prior work is that it does not model a mechanism's uncertainty over values for not yet elicited bundles. In this paper, we address this shortcoming by presenting a Bayesian optimization-based combinatorial assignment (BOCA) mechanism. Our key technical contribution is to integrate a method for capturing model uncertainty into an iterative combinatorial auction mechanism. Concretely, we design a new method for estimating an upper uncertainty bound that can be used to define an acquisition function to determine the next query to the agents. This enables the mechanism to properly explore (and not just exploit) the bundle space during its preference elicitation phase. We run computational experiments in several spectrum auction domains to evaluate BOCA's performance. Our results show that BOCA achieves higher allocative efficiency than state-of-the-art approaches.

We propose a model-driven decision support system (DSS) based on a Bayesian belief network (BBN) to support cyber deception based on a detailed model of attacker beliefs. We discuss this approach using a case study based on passively observed operating system (OS) fingerprinting data. In passive reconnaissance attackers can remain undetected while collecting information to identify systems and plan attacks. Our DSS is intended to support preventative measures to protect the network from successful reconnaissance, such as by modifying features using deception. We validate the prediction accuracy of the model in comparison with a sequential artificial neural network (ANN). We then introduce a deceptive algorithm to select a minimal set of features for OS obfuscation. We show the effectiveness of feature-modification strategies based on our methods using passively collected data to decide what features from a real operating system (OS) to modify to appear as a fake [different] OS.

Federated learning is a contemporary machine learning paradigm where locally trained models are distilled into a global model. Due to the intrinsic permutation invariance of neural networks, Probabilistic Federated Neural Matching (PFNM) employs a Bayesian nonparametric framework in the generation process of local neurons, and then creates a linear sum assignment formulation in each alternative optimization iteration. But according to our theoretical analysis, the optimization iteration in PFNM omits global information from existing. In this study, we propose a novel approach that overcomes this flaw by introducing a Kullback-Leibler divergence penalty at each iteration. The effectiveness of our approach is demonstrated by experiments on both image classification and semantic segmentation tasks.

Recent advances in large pre-trained models showed promising results in few-shot learning. However, their generalization ability on two-dimensional Out-of-Distribution (OoD) data, i.e., correlation shift and diversity shift, has not been thoroughly investigated. Researches have shown that even with a significant amount of training data, few methods can achieve better performance than the standard empirical risk minimization method (ERM) in OoD generalization. This few-shot OoD generalization dilemma emerges as a challenging direction in deep neural network generalization research, where the performance suffers from overfitting on few-shot examples and OoD generalization errors. In this paper, leveraging a broader supervision source, we explore a novel Bayesian cross-modal image-text alignment learning method (Bayes-CAL) to address this issue. Specifically, the model is designed as only text representations are fine-tuned via a Bayesian modelling approach with gradient orthogonalization loss and invariant risk minimization (IRM) loss. The Bayesian approach is essentially introduced to avoid overfitting the base classes observed during training and improve generalization to broader unseen classes. The dedicated loss is introduced to achieve better image-text alignment by disentangling the causal and non-casual parts of image features. Numerical experiments demonstrate that Bayes-CAL achieved state-of-the-art OoD generalization performances on two-dimensional distribution shifts. Moreover, compared with CLIP-like models, Bayes-CAL yields more stable generalization performances on unseen classes. Our code is available at https://github.com/LinLLLL/BayesCAL.

Instant runoff voting (IRV) is an increasingly-popular alternative to traditional plurality voting in which voters submit rankings over the candidates rather than single votes. In practice, elections using IRV often restrict the ballot length, the number of candidates a voter is allowed to rank on their ballot. We theoretically and empirically analyze how ballot length can influence the outcome of an election, given fixed voter preferences. We show that there exist preference profiles over k candidates such that up to k-1 different candidates win at different ballot lengths. We derive exact lower bounds on the number of voters required for such profiles and provide a construction matching the lower bound for unrestricted voter preferences. Additionally, we characterize which sequences of winners are possible over ballot lengths and provide explicit profile constructions achieving any feasible winner sequence. We also examine how classic preference restrictions influence our results—for instance, single-peakedness makes k-1 different winners impossible but still allows at least Ω(√k). Finally, we analyze a collection of 168 real-world elections, where we truncate rankings to simulate shorter ballots. We find that shorter ballots could have changed the outcome in one quarter of these elections. Our results highlight ballot length as a consequential degree of freedom in the design of IRV elections.

In real-world scenarios, it is crucial to build a lifelong taskoriented dialogue system (TDS) that continually adapts to new knowledge without forgetting previously acquired experiences. Existing approaches mainly focus on mitigating the catastrophic forgetting in lifelong TDS. However, the transfer ability to generalize the accumulated old knowledge to new tasks is underexplored. In this paper, we propose a two-stage lifelong task-oriented dialogue generation method to mitigate catastrophic forgetting and encourage knowledge transfer simultaneously, inspired by the learning process. In the first stage, we learn task-specific masks which adaptively preserve the knowledge of each visited task so as to mitigate catastrophic forgetting. In this stage, we are expected to learn the task-specific knowledge which is tailored for each task. In the second stage, we bring the knowledge from the encountered tasks together and understand thoroughly. To this end, we devise a balanced meta learning strategy for both forward and backward knowledge transfer in the lifelong learning process. In particular, we perform meta-update with a meta-test set sampled from the current training data for forward knowledge transfer. In addition, we employ an uncertainty-based sampling strategy to select and store representative dialogue samples into episodic memory and perform meta-update with a meta-test set sampled from the memory for backward knowledge transfer. With extensive experiments on 29 tasks, we show that MetaLTDS outperforms the strong baselines in terms of both effectiveness and efficiency. For reproducibility, we submit our code at: https: //github.com/travis-xu/MetaLTDS.

Pruning has been an effective solution to reduce the number of computations and the memory requirement in deep learning. The pruning unit plays an important role in exploiting the GPU resources efficiently. The filter is proposed as a simple pruning unit of structured pruning. However, since the filter is quite large as pruning unit, the accuracy drop is considerable with a high pruning ratio. GPU rearranges the weight and input tensors into tiles (blocks) for efficient computation. To fully utilize GPU resources, this tile structure should be considered, which is the goal of block pruning. However, previous block pruning prunes both row vectors and column vectors. Pruning of row vectors in a tile corresponds to filter pruning, and it also interferes with column-wise block pruning of the following layer. In contrast, column vectors are much smaller than row vectors and can achieve lower accuracy drop. Additionally, if the pruning ratio for each tile is different, GPU utilization can be limited by imbalanced workloads by irregular-sized blocks. The same pruning ratio for the weight tiles processed in parallel enables the actual inference process to fully utilize the resources without idle time. This paper proposes balanced column-wise block pruning, named BCBP, to satisfy two conditions: the column-wise minimal size of the pruning unit and balanced workloads. We demonstrate that BCBP is superior to previous pruning methods through comprehensive experiments.

Training deep neural networks on large-scale datasets requires significant hardware resources whose costs (even on cloud platforms) put them out of reach of smaller organizations, groups, and individuals. Backpropagation (backprop), the workhorse for training these networks, is an inherently sequential process that is difficult to parallelize. Furthermore, researchers must continually develop various specialized techniques, such as particular weight initializations and enhanced activation functions, to ensure stable parameter optimization. Our goal is to seek an effective, neuro-biologically plausible alternative to backprop that can be used to train deep networks. In this paper, we propose a backprop-free procedure, recursive local representation alignment, for training large-scale architectures. Experiments with residual networks on CIFAR-10 and the large benchmark, ImageNet, show that our algorithm generalizes as well as backprop while converging sooner due to weight updates that are parallelizable and computationally less demanding. This is empirical evidence that a backprop-free algorithm can scale up to larger datasets.

Change detection (CD) is to decouple object changes (i.e., object missing or appearing) from background changes (i.e., environment variations) like light and season variations in two images captured in the same scene over a long time span, presenting critical applications in disaster management, urban development, etc. In particular, the endless patterns of background changes require detectors to have a high generalization against unseen environment variations, making this task significantly challenging. Recent deep learning-based methods develop novel network architectures or optimization strategies with paired-training examples, which do not handle the generalization issue explicitly and require huge manual pixel-level annotation efforts. In this work, for the first attempt in the CD community, we study the generalization issue of CD from the perspective of data augmentation and develop a novel weakly supervised training algorithm that only needs image-level labels. Different from general augmentation techniques for classification, we propose the background-mixed augmentation that is specifically designed for change detection by augmenting examples under the guidance of a set of background changing images and letting deep CD models see diverse environment variations. Moreover, we propose the augmented & real data consistency loss that encourages the generalization increase significantly. Our method as a general framework can enhance a wide range of existing deep learning-based detectors. We conduct extensive experiments in two public datasets and enhance four state-of-the-art methods, demonstrating the advantages of our method. We release the code at https://github.com/tsingqguo/bgmix.

In this paper we introduce Backforward Propagation, a method of completely eliminating Internal Covariate Shift (ICS). Unlike previous methods, which only indirectly reduce the impact of ICS while introducing other biases, we are able to have a surgical view at the effects ICS has on training neural networks. Our experiments show that ICS has a weight regularizing effect on models, and completely removing it enables for faster convergence of the neural network.

State of the art methods for ad hoc teamwork, i.e., for collaboration without prior coordination, often use a long history of prior observations to model the behavior of other agents (or agent types) and to determine the ad hoc agent's behavior. In many practical domains, it is difficult to obtain large training datasets, and necessary to quickly revise the existing models to account for changes in team composition or domain attributes. Our architecture builds on the principles of step-wise refinement and ecological rationality to enable an ad hoc agent to perform non-monotonic logical reasoning with prior commonsense domain knowledge and models learned rapidly from limited examples to predict the behavior of other agents. In the simulated multiagent collaboration domain Fort Attack, we experimentally demonstrate that our architecture enables an ad hoc agent to adapt to changes in the behavior of other agents, and provides enhanced transparency and better performance than a state of the art data-driven baseline.

The partially occluded image recognition (POIR) problem has been a challenge for artificial intelligence for a long time. A common strategy to handle the POIR problem is using the non-occluded features for classification. Unfortunately, this strategy will lose effectiveness when the image is severely occluded, since the visible parts can only provide limited information. Several studies in neuroscience reveal that feature restoration which fills in the occluded information and is called amodal completion is essential for human brains to recognize partially occluded images. However, feature restoration is commonly ignored by CNNs, which may be the reason why CNNs are ineffective for the POIR problem. Inspired by this, we propose a novel brain-inspired feature restoration network (BIFRNet) to solve the POIR problem. It mimics a ventral visual pathway to extract image features and a dorsal visual pathway to distinguish occluded and visible image regions. In addition, it also uses a knowledge module to store classification prior knowledge and uses a completion module to restore occluded features based on visible features and prior knowledge. Thorough experiments on synthetic and real-world occluded image datasets show that BIFRNet outperforms the existing methods in solving the POIR problem. Especially for severely occluded images, BIRFRNet surpasses other methods by a large margin and is close to the human brain performance. Furthermore, the brain-inspired design makes BIFRNet more interpretable.

Restricted by the ability of depth perception, all Multi-view 3D object detection methods fall into the bottleneck of depth accuracy. By constructing temporal stereo, depth estimation is quite reliable in indoor scenarios. However, there are two difficulties in directly integrating temporal stereo into outdoor multi-view 3D object detectors: 1) The construction of temporal stereos for all views results in high computing costs. 2) Unable to adapt to challenging outdoor scenarios. In this study, we propose an effective method for creating temporal stereo by dynamically determining the center and range of the temporal stereo. The most confident center is found using the EM algorithm. Numerous experiments on nuScenes have shown the BEVStereo's ability to deal with complex outdoor scenarios that other stereo-based methods are unable to handle. For the first time, a stereo-based approach shows superiority in scenarios like a static ego vehicle and moving objects. BEVStereo achieves the new state-of-the-art in the camera-only track of nuScenes dataset while maintaining memory efficiency. Codes have been released.