AAAI 2023 论文列表

Brain-computer Interface (BCI) builds a neural signal to the motor command pathway, which is a prerequisite for the realization of neural prosthetics. However, a long-term stable BCI suffers from the neural data drift across days while retraining the BCI decoder is expensive and restricts its application scenarios. Recent solutions of neural signal recalibration treat the continuous neural signals as discrete, which is less effective in temporal feature extraction. Inspired by the observation from biologists that low-dimensional dynamics could describe high-dimensional neural signals, we model the underlying neural dynamics and propose a semantic-dynamic feature that represents the semantics and dynamics in a shared feature space facilitating the BCI recalibration. Besides, we present the joint distribution alignment instead of the common used marginal alignment strategy, dealing with the various complex changes in neural data distribution. Our recalibration approach achieves state-of-the-art performance on the real neural data of two monkeys in both classification and regression tasks. Our approach is also evaluated on a simulated dataset, which indicates its robustness in dealing with various common causes of neural signal instability.

Recent years have witnessed the great success of Graph Neural Networks (GNNs) in handling graph-related tasks. However, MLPs remain the primary workhorse for practical industrial applications due to their desirable inference efficiency and scalability. To reduce their gaps, one can directly distill knowledge from a well-designed teacher GNN to a student MLP, which is termed as GNN-to-MLP distillation. However, the process of distillation usually entails a loss of information, and ``which knowledge patterns of GNNs are more likely to be left and distilled into MLPs?" becomes an important question. In this paper, we first factorize the knowledge learned by GNNs into low- and high-frequency components in the spectral domain and then derive their correspondence in the spatial domain. Furthermore, we identified a potential information drowning problem for existing GNN-to-MLP distillation, i.e., the high-frequency knowledge of the pre-trained GNNs may be overwhelmed by the low-frequency knowledge during distillation; we have described in detail what it represents, how it arises, what impact it has, and how to deal with it. In this paper, we propose an efficient Full-Frequency GNN-to-MLP (FF-G2M) distillation framework, which extracts both low-frequency and high-frequency knowledge from GNNs and injects it into MLPs. Extensive experiments show that FF-G2M improves over the vanilla MLPs by 12.6% and outperforms its corresponding teacher GNNs by 2.6% averaged over six graph datasets and three common GNN architectures.

Temporal Planning is the problem of synthesizing a course of actions given a predictive model of a system subject to temporal constraints. This kind of planning finds natural applications in the automation of industrial processes and in robotics when the timing and deadlines are important. Finding any plan in temporal planning is often not enough as it is sometimes needed to optimize a certain objective function: particularly interesting are the minimization of the makespan and the optimization of the costs of actions. Despite the importance of the problem, only few works in the literature tackled the problem of optimal temporal planning because of the complicated intermix of planning and scheduling. In this paper, we address the problem of optimal temporal planning for a very expressive class of problems using a reduction of the bounded planning problem to Optimization Modulo Theory (OMT) a powerful discrete/continuous optimization framework. We theoretically and empirically show the expressive power of this approach and we set a baseline for future research in this area.

Self-supervised space-time correspondence learning is emerging as a promising way of leveraging unlabeled video. Currently, most methods adapt contrastive learning with mining negative samples or reconstruction adapted from the image domain, which requires dense affinity across multiple frames or optical flow constraints. Moreover, video correspondence predictive models require mining more inherent properties in videos, such as structural information. In this work, we propose the VideoHiGraph, a space-time correspondence framework based on a learnable graph kernel. Concerning the video as the spatial-temporal graph, the learning objectives of VideoHiGraph are emanated in a self-supervised manner for predicting unobserved hidden graphs via graph kernel manner. We learn a representation of the temporal coherence across frames in which pairwise similarity defines the structured hidden graph, such that a biased random walk graph kernel along the sub-graph can predict long-range correspondence. Then, we learn a refined representation across frames on the node-level via a dense graph kernel. The self-supervision of the model training is formed by the structural and temporal consistency of the graph. VideoHiGraph achieves superior performance and demonstrates its robustness across the benchmark of label propagation tasks involving objects, semantic parts, keypoints, and instances. Our algorithm implementations have been made publicly available at https://github.com/zyqin19/VideoHiGraph.

Compression techniques in machine learning (ML) independently improve a model’s inference efficiency by reducing its memory footprint while aiming to maintain its quality. This paper lays groundwork in questioning the merit of a compression pipeline involving all techniques as opposed to skipping a few by considering a case study on a keyword spotting model: DS-CNN-S. In addition, it documents improvements to the model’s training and dataset infrastructure. For this model, preliminary findings suggest that a full-scale pipeline isn’t required to achieve a competent memory footprint and accuracy, but a more comprehensive study is required.

The most recent approaches for clustering single-cell RNA-sequencing data rely on deep auto-encoders. However, three major challenges remain unaddressed. First, current models overlook the impact of the cumulative errors induced by the pseudo-supervised embedding clustering task (Feature Randomness). Second, existing methods neglect the effect of the strong competition between embedding clustering and reconstruction (Feature Drift). Third, the previous deep clustering models regularly fail to consider the topological information of the latent data, even though the local and global latent configurations can bring complementary views to the clustering task. To address these challenges, we propose a novel approach that explores the interaction between local and global latent configurations to progressively adjust the reconstruction and embedding clustering tasks. We elaborate a topological and probabilistic filter to mitigate Feature Randomness and a cell-cell graph structure and content correction mechanism to counteract Feature Drift. The Zero-Inflated Negative Binomial model is also integrated to capture the characteristics of gene expression profiles. We conduct detailed experiments on real-world datasets from multiple representative genome sequencing platforms. Our approach outperforms the state-of-the-art clustering methods in various evaluation metrics.

The success of Deep Generative Models at high-resolution image generation has led to their extensive utilization for style editing of real images. Most existing methods work on the principle of inverting real images onto their latent space, followed by determining controllable directions. Both inversion of real images and determination of controllable latent directions are computationally expensive operations. Moreover, the determination of controllable latent directions requires additional human supervision. This work aims to explore the efficacy of mask-guided feature modulation in the latent space of a Deep Generative Model as a solution to these bottlenecks. To this end, we present the SemanticStyle Autoencoder (SSAE), a deep Generative Autoencoder model that leverages semantic mask-guided latent space manipulation for highly localized photorealistic style editing of real images. We present qualitative and quantitative results for the same and their analysis. This work shall serve as a guiding primer for future work.

Human has the remarkable ability of learning novel objects by browsing extremely few examples, which may be attributed to the generic and robust feature extracted in the ventral stream of our brain for representing visual objects. In this sense, the tuning characteristics of ventral stream's neurons can be useful prior knowledge to improve few-shot classification. Specifically, we computationally model two groups of neurons found in ventral stream which are respectively sensitive to shape cues and color cues. Then we propose the hierarchical feature regularization method with these neuron models to regularize the backbone of a few-shot model, thus making it produce more generic and robust features for few-shot classification. In addition, to simulate the tuning characteristic that neuron firing at a higher rate in response to foreground stimulus elements compared to background elements, which we call belongingness, we design a foreground segmentation algorithm based on the observation that the foreground object usually does not appear at the edge of the picture, then multiply the foreground mask with the backbone of few-shot model. Our method is model-agnostic and can be applied to few-shot models with different backbones, training paradigms and classifiers.

The US public school system is administered by local school districts. Each district comprises a set of schools mapped to attendance zones which are annually assessed to meet enrollment objectives. To support school officials in redrawing attendance boundaries, existing approaches have proven promising but still suffer from several challenges, including: 1) inability to scale to large school districts, 2) high computational cost of obtaining compact school attendance zones, and 3) lack of discussion on quantifying ethical considerations underlying the redrawing of school boundaries. Motivated by these challenges, this paper approaches the school redistricting problem from both computational and ethical standpoints. First, we introduce a practical framework based on sampling methods to solve school redistricting as a graph partitioning problem. Next, the advantages of adopting a modified objective function for optimizing discrete geometry to obtain compact boundaries are examined. Lastly, alternative metrics to address ethical considerations in real-world scenarios are formally defined and thoroughly discussed. Our findings highlight the inclusiveness and efficiency advantages of the designed framework and depict how tradeoffs need to be made to obtain qualitatively different school redistricting plans.

Most existing Spiking Neural Network (SNN) works state that SNNs may utilize temporal information dynamics of spikes. However, an explicit analysis of temporal information dynamics is still missing. In this paper, we ask several important questions for providing a fundamental understanding of SNNs: What are temporal information dynamics inside SNNs? How can we measure the temporal information dynamics? How do the temporal information dynamics affect the overall learning performance? To answer these questions, we estimate the Fisher Information of the weights to measure the distribution of temporal information during training in an empirical manner. Surprisingly, as training goes on, Fisher information starts to concentrate in the early timesteps. After training, we observe that information becomes highly concentrated in earlier few timesteps, a phenomenon we refer to as temporal information concentration. We observe that the temporal information concentration phenomenon is a common learning feature of SNNs by conducting extensive experiments on various configurations such as architecture, dataset, optimization strategy, time constant, and timesteps. Furthermore, to reveal how temporal information concentration affects the performance of SNNs, we design a loss function to change the trend of temporal information. We find that temporal information concentration is crucial to building a robust SNN but has little effect on classification accuracy. Finally, we propose an efficient iterative pruning method based on our observation on temporal information concentration. Code is available at https://github.com/Intelligent-Computing-Lab-Yale/Exploring-Temporal-Information-Dynamics-in-Spiking-Neural-Networks.

Scene text editing (STE) aims to replace text with the desired one while preserving background and styles of the original text. However, due to the complicated background textures and various text styles, existing methods fall short in generating clear and legible edited text images. In this study, we attribute the poor editing performance to two problems: 1) Implicit decoupling structure. Previous methods of editing the whole image have to learn different translation rules of background and text regions simultaneously. 2) Domain gap. Due to the lack of edited real scene text images, the network can only be well trained on synthetic pairs and performs poorly on real-world images. To handle the above problems, we propose a novel network by MOdifying Scene Text image at strokE Level (MOSTEL). Firstly, we generate stroke guidance maps to explicitly indicate regions to be edited. Different from the implicit one by directly modifying all the pixels at image level, such explicit instructions filter out the distractions from background and guide the network to focus on editing rules of text regions. Secondly, we propose a Semi-supervised Hybrid Learning to train the network with both labeled synthetic images and unpaired real scene text images. Thus, the STE model is adapted to real-world datasets distributions. Moreover, two new datasets (Tamper-Syn2k and Tamper-Scene) are proposed to fill the blank of public evaluation datasets. Extensive experiments demonstrate that our MOSTEL outperforms previous methods both qualitatively and quantitatively. Datasets and code will be available at https://github.com/qqqyd/MOSTEL.

Autoregressive language modeling (ALM) has been successfully used in self-supervised pre-training in Natural language processing (NLP). However, this paradigm has not achieved comparable results with other self-supervised approaches in computer vision (e.g., contrastive learning, masked image modeling). In this paper, we try to find the reason why autoregressive modeling does not work well on vision tasks. To tackle this problem, we fully analyze the limitation of visual autoregressive methods and proposed a novel stochastic autoregressive image modeling (named SAIM) by the two simple designs. First, we serialize the image into patches. Second, we employ the stochastic permutation strategy to generate an effective and robust image context which is critical for vision tasks. To realize this task, we create a parallel encoder-decoder training process in which the encoder serves a similar role to the standard vision transformer focusing on learning the whole contextual information, and meanwhile the decoder predicts the content of the current position so that the encoder and decoder can reinforce each other. Our method significantly improves the performance of autoregressive image modeling and achieves the best accuracy (83.9%) on the vanilla ViT-Base model among methods using only ImageNet-1K data. Transfer performance in downstream tasks also shows that our model achieves competitive performance. Code is available at https://github.com/qiy20/SAIM.

Large neural network-based language models play an increasingly important role in contemporary AI. Although these models demonstrate sophisticated text generation capabilities, they have also been shown to reproduce harmful social biases contained in their training data. This paper presents a project that guides students through an exploration of social biases in large language models. As a final project for an intermediate college course in Artificial Intelligence, students developed a bias probe task for a previously-unstudied aspect of sociolinguistic or sociocultural bias they were interested in exploring. Through the process of constructing a dataset and evaluation metric to measure bias, students mastered key technical concepts, including how to run contemporary neural networks for natural language processing tasks; construct datasets and evaluation metrics; and analyze experimental results. Students reported their findings in an in-class presentation and a final report, recounting patterns of predictions that surprised, unsettled, and sparked interest in advocating for technology that reflects a more diverse set of backgrounds and experiences. Through this project, students engage with and even contribute to a growing body of scholarly work on social biases in large language models.

Document-level relation extraction (RE) aims to extract relational triples from a document. One of its primary challenges is to predict implicit relations between entities, which are not explicitly expressed in the document but can usually be extracted through relational reasoning. Previous methods mainly implicitly model relational reasoning through the interaction among entities or entity pairs. However, they suffer from two deficiencies: 1) they often consider only one reasoning pattern, of which coverage on relational triples is limited; 2) they do not explicitly model the process of relational reasoning. In this paper, to deal with the first problem, we propose a document-level RE model with a reasoning module that contains a core unit, the reasoning multi-head self-attention unit. This unit is a variant of the conventional multi-head self-attention and utilizes four attention heads to model four common reasoning patterns, respectively, which can cover more relational triples than previous methods. Then, to address the second issue, we propose a self-distillation training framework, which contains two branches sharing parameters. In the first branch, we first randomly mask some entity pair feature vectors in the document, and then train our reasoning module to infer their relations by exploiting the feature information of other related entity pairs. By doing so, we can explicitly model the process of relational reasoning. However, because the additional masking operation is not used during testing, it causes an input gap between training and testing scenarios, which would hurt the model performance. To reduce this gap, we perform conventional supervised training without masking operation in the second branch and utilize Kullback-Leibler divergence loss to minimize the difference between the predictions of the two branches. Finally, we conduct comprehensive experiments on three benchmark datasets, of which experimental results demonstrate that our model consistently outperforms all competitive baselines. Our source code is available at https://github.com/DeepLearnXMU/DocRE-SD

The ensemble attack with average weights can be leveraged for increasing the transferability of universal adversarial perturbation (UAP) by training with multiple Convolutional Neural Networks (CNNs). However, after analyzing the Pearson Correlation Coefficients (PCCs) between the ensemble logits and individual logits of the crafted UAP trained by the ensemble attack, we find that one CNN plays a dominant role during the optimization. Consequently, this average weighted strategy will weaken the contributions of other CNNs and thus limit the transferability for other black-box CNNs. To deal with this bias issue, the primary attempt is to leverage the Kullback–Leibler (KL) divergence loss to encourage the joint contribution from different CNNs, which is still insufficient. After decoupling the KL loss into a target-class part and a non-target-class part, the main issue lies in that the non-target knowledge will be significantly suppressed due to the increasing logit of the target class. In this study, we simply adopt a KL loss that only considers the non-target classes for addressing the dominant bias issue. Besides, to further boost the transferability, we incorporate the min-max learning framework to self-adjust the ensemble weights for each CNN. Experiments results validate that considering the non-target KL loss can achieve superior transferability than the original KL loss by a large margin, and the min-max training can provide a mutual benefit in adversarial ensemble attacks. The source code is available at: https://github.com/WJJLL/ND-MM.

In financial economics, studies have shown that the textual content in the earnings conference call transcript has predictive power for a firm's future risk. However, the conference call transcript is very long and contains diverse non-relevant content, which poses challenges for the text-based risk forecast. This study investigates the structural dependency within a conference call transcript by explicitly modeling the dialogue between managers and analysts. Specifically, we utilize TextRank to extract information and exploit the semantic correlation within a discussion using hypergraph learning. This novel design can improve the transcript representation performance and reduce the risk of forecast errors. Experimental results on a large-scale dataset show that our approach can significantly improve prediction performance compared to state-of-the-art text-based models.

The opaqueness of the multi-hop fact verification model imposes imperative requirements for explainability. One feasible way is to extract rationales, a subset of inputs, where the performance of prediction drops dramatically when being removed. Though being explainable, most rationale extraction methods for multi-hop fact verification explore the semantic information within each piece of evidence individually, while ignoring the topological information interaction among different pieces of evidence. Intuitively, a faithful rationale bears complementary information being able to extract other rationales through the multi-hop reasoning process. To tackle such disadvantages, we cast explainable multi-hop fact verification as subgraph extraction, which can be solved based on graph convolutional network (GCN) with salience-aware graph learning. In specific, GCN is utilized to incorporate the topological interaction information among multiple pieces of evidence for learning evidence representation. Meanwhile, to alleviate the influence of noisy evidence, the salience-aware graph perturbation is induced into the message passing of GCN. Moreover, the multi-task model with three diagnostic properties of rationale is elaborately designed to improve the quality of an explanation without any explicit annotations. Experimental results on the FEVEROUS benchmark show significant gains over previous state-of-the-art methods for both rationale extraction and fact verification.

Measuring the perception of visual content is a long-standing problem in computer vision. Many mathematical models have been developed to evaluate the look or quality of an image. Despite the effectiveness of such tools in quantifying degradations such as noise and blurriness levels, such quantification is loosely coupled with human language. When it comes to more abstract perception about the feel of visual content, existing methods can only rely on supervised models that are explicitly trained with labeled data collected via laborious user study. In this paper, we go beyond the conventional paradigms by exploring the rich visual language prior encapsulated in Contrastive Language-Image Pre-training (CLIP) models for assessing both the quality perception (look) and abstract perception (feel) of images without explicit task-specific training. In particular, we discuss effective prompt designs and show an effective prompt pairing strategy to harness the prior. We also provide extensive experiments on controlled datasets and Image Quality Assessment (IQA) benchmarks. Our results show that CLIP captures meaningful priors that generalize well to different perceptual assessments.

Exploring Artificial Intelligence (AI) in English Language Arts (ELA) with StoryQ is a 10-hour curriculum module designed for high school ELA classes. The module introduces students to fundamental AI concepts and essential machine learning workflow using StoryQ, a web-based GUI environment for Grades 6-12 learners. In this module, students work with unstructured text data and learn to train, test, and improve text classification models such as intent recognition, clickbait filter, and sentiment analysis. As they interact with machine-learning language models deeply, students also gain a nuanced understanding of language and how to wield it, not just as a data structure, but as a tool in our human-human encounters as well. The current version contains eight lessons, all delivered through a full-featured online learning and teaching platform. Computers and Internet access are required to implement the module. The module was piloted in an ELA class in the Spring of 2022, and the student learning outcomes were positive. The module is currently undergoing revision and will be further tested and improved in Fall 2022.