AAAI 2023 论文列表

Multivariate time series classification (MTSC), one of the most fundamental time series applications, has not only gained substantial research attentions but has also emerged in many real-life applications. Recently, using transformers to solve MTSC has been reported. However, current transformer-based methods take data points of individual timestamps as inputs (timestamp-level), which only capture the temporal dependencies, not the dependencies among variables. In this paper, we propose a novel method, called SVP-T. Specifically, we first propose to take time series subsequences, which can be from different variables and positions (time interval), as the inputs (shape-level). The temporal and variable dependencies are both handled by capturing the long- and short-term dependencies among shapes. Second, we propose a variable-position encoding layer (VP-layer) to utilize both the variable and position information of each shape. Third, we introduce a novel VP-based (Variable-Position) self-attention mechanism to allow the enhancing the attention weights of overlapping shapes. We evaluate our method on all UEA MTS datasets. SVP-T achieves the best accuracy rank when compared with several competitive state-of-the-art methods. Furthermore, we demonstrate the effectiveness of the VP-layer and the VP-based self-attention mechanism. Finally, we present one case study to interpret the result of SVP-T.

Occlusion and motion blur make it challenging to interpolate video frame, since estimating complex motions between two frames is hard and unreliable, especially in highly dynamic scenes. This paper aims to address these issues by exploiting spike stream as auxiliary visual information between frames to synthesize target frames. Instead of estimating motions by optical flow from RGB frames, we present a new dual-modal pipeline adopting both RGB frames and the corresponding spike stream as inputs (SVFI). It extracts the scene structure and objects' outline feature maps of the target frames from spike stream. Those feature maps are fused with the color and texture feature maps extracted from RGB frames to synthesize target frames. Benefited by the spike stream that contains consecutive information between two frames, SVFI can directly extract the information in occlusion and motion blur areas of target frames from spike stream, thus it is more robust than previous optical flow-based methods. Experiments show SVFI outperforms the SOTA methods on wide variety of datasets. For instance, in 7 and 15 frame skip evaluations, it shows up to 5.58 dB and 6.56 dB improvements in terms of PSNR over the corresponding second best methods BMBC and DAIN. SVFI also shows visually impressive performance in real-world scenes.

Although large-scale video-language pre-training models, which usually build a global alignment between the video and the text, have achieved remarkable progress on various downstream tasks, the idea of adopting fine-grained information during the pre-training stage is not well explored. In this work, we propose STOA-VLP, a pre-training framework that jointly models object and action information across spatial and temporal dimensions. More specifically, the model regards object trajectories across frames and multiple action features from the video as fine-grained features. Besides, We design two auxiliary tasks to better incorporate both kinds of information into the pre-training process of the video-language model. The first is the dynamic object-text alignment task, which builds a better connection between object trajectories and the relevant noun tokens. The second is the spatial-temporal action set prediction, which guides the model to generate consistent action features by predicting actions found in the text. Extensive experiments on three downstream tasks (video captioning, text-video retrieval, and video question answering) demonstrate the effectiveness of our proposed STOA-VLP (e.g. 3.7 Rouge-L improvements on MSR-VTT video captioning benchmark, 2.9% accuracy improvements on MSVD video question answering benchmark, compared to previous approaches).

Deep Reinforcement Learning (DRL) has the potential to be used for synthesizing feedback controllers (agents) for various complex systems with unknown dynamics. These systems are expected to satisfy diverse safety and liveness properties best captured using temporal logic. In RL, the reward function plays a crucial role in specifying the desired behaviour of these agents. However, the problem of designing the reward function for an RL agent to satisfy complex temporal logic specifications has received limited attention in the literature. To address this, we provide a systematic way of generating rewards in real-time by using the quantitative semantics of Signal Temporal Logic (STL), a widely used temporal logic to specify the behaviour of cyber-physical systems. We propose a new quantitative semantics for STL having several desirable properties, making it suitable for reward generation. We evaluate our STL-based reinforcement learning mechanism on several complex continuous control benchmarks and compare our STL semantics with those available in the literature in terms of their efficacy in synthesizing the controller agent. Experimental results establish our new semantics to be the most suitable for synthesizing feedback controllers for complex continuous dynamical systems through reinforcement learning.

Active learning (AL) aims to sample the most informative data instances for labeling, which makes the model fitting data efficient while significantly reducing the annotation cost. However, most existing AL models make a strong assumption that the annotated data instances are always assigned correct labels, which may not hold true in many practical settings. In this paper, we develop a theoretical framework to formally analyze the impact of noisy annotations and show that systematically re-sampling guarantees to reduce the noise rate, which can lead to improved generalization capability. More importantly, the theoretical framework demonstrates the key benefit of conducting active re-sampling on label-efficient learning, which is critical for AL. The theoretical results also suggest essential properties of an active re-sampling function with a fast convergence speed and guaranteed error reduction. This inspires us to design a novel spatial-temporal active re-sampling function by leveraging the important spatial and temporal properties of maximum-margin classifiers. Extensive experiments conducted on both synthetic and real-world data clearly demonstrate the effectiveness of the proposed active re-sampling function.

Aspect Sentiment Triplet Extraction (ASTE) has become an emerging task in sentiment analysis research, aiming to extract triplets of the aspect term, its corresponding opinion term, and its associated sentiment polarity from a given sentence. Recently, many neural networks based models with different tagging schemes have been proposed, but almost all of them have their limitations: heavily relying on 1) prior assumption that each word is only associated with a single role (e.g., aspect term, or opinion term, etc. ) and 2) word-level interactions and treating each opinion/aspect as a set of independent words. Hence, they perform poorly on the complex ASTE task, such as a word associated with multiple roles or an aspect/opinion term with multiple words. Hence, we propose a novel approach, Span TAgging and Greedy infErence (STAGE), to extract sentiment triplets in span-level, where each span may consist of multiple words and play different roles simultaneously. To this end, this paper formulates the ASTE task as a multi-class span classification problem. Specifically, STAGE generates more accurate aspect sentiment triplet extractions via exploring span-level information and constraints, which consists of two components, namely, span tagging scheme and greedy inference strategy. The former tag all possible candidate spans based on a newly-defined tagging set. The latter retrieves the aspect/opinion term with the maximum length from the candidate sentiment snippet to output sentiment triplets. Furthermore, we propose a simple but effective model based on the STAGE, which outperforms the state-of-the-arts by a large margin on four widely-used datasets. Moreover, our STAGE can be easily generalized to other pair/triplet extraction tasks, which also demonstrates the superiority of the proposed scheme STAGE.

Hard-label textual adversarial attack is a challenging task, as only the predicted label information is available, and the text space is discrete and non-differentiable. Relevant research work is still in fancy and just a handful of methods are proposed. However, existing methods suffer from either the high complexity of genetic algorithms or inaccurate gradient estimation, thus are arduous to obtain adversarial examples with high semantic similarity and low perturbation rate under the tight-budget scenario. In this paper, we propose a simple and sweet paradigm for hard-label textual adversarial attack, named SSPAttack. Specifically, SSPAttack first utilizes initialization to generate an adversarial example, and removes unnecessary replacement words to reduce the number of changed words. Then it determines the replacement order and searches for an anchor synonym, thus avoiding going through all the synonyms. Finally, it pushes substitution words towards original words until an appropriate adversarial example is obtained. The core idea of SSPAttack is just swapping words whose mechanism is simple. Experimental results on eight benchmark datasets and two real-world APIs have shown that the performance of SSPAttack is sweet in terms of similarity, perturbation rate and query efficiency.

The Chinese NER task consists of two steps, first determining entity boundaries and then labeling them. Some previous work incorporating related words from pre-trained vocabulary into character-based models has been demonstrated to be effective. However, the number of words that characters can match in the vocabulary is large, and their meanings vary widely. It is unreasonable to concatenate all the matched words into the character's representation without making semantic distinctions. This is because words with different semantics also have distinct vectors by the distributed representation. Moreover, mutual information maximization (MIM) provides a unified way to characterize the correction between different granularity of embeddings, we find it can be used to enhance the features in our task. Consequently, this paper introduces a novel Chinese NER model named SSMI based on semantic similarity and MIM. We first match all the potential word boundaries of the input characters from the pre-trained vocabulary and employ BERT to segment the input sentence to get the segmentation containing these characters. After computing their cosine similarity, we obtain the word boundary with the highest similarity and the word group with similarity score larger than a specific threshold. Then, we concatenate the most relevant word boundaries with character vectors. We further calculate the mutual information maximization of group, character and sentence, respectively. Finally, we feed the result from the above steps to our novel network. The results on four Chinese public NER datasets show that our SSMI achieves state-of-the-art performance.

LiDAR-based 3D object detection is an indispensable task in advanced autonomous driving systems. Though impressive detection results have been achieved by superior 3D detectors, they suffer from significant performance degeneration when facing unseen domains, such as different LiDAR configurations, different cities, and weather conditions. The mainstream approaches tend to solve these challenges by leveraging unsupervised domain adaptation (UDA) techniques. However, these UDA solutions just yield unsatisfactory 3D detection results when there is a severe domain shift, e.g., from Waymo (64-beam) to nuScenes (32-beam). To address this, we present a novel Semi-Supervised Domain Adaptation method for 3D object detection (SSDA3D), where only a few labeled target data is available, yet can significantly improve the adaptation performance. In particular, our SSDA3D includes an Inter-domain Adaptation stage and an Intra-domain Generalization stage. In the first stage, an Inter-domain Point-CutMix module is presented to efficiently align the point cloud distribution across domains. The Point-CutMix generates mixed samples of an intermediate domain, thus encouraging to learn domain-invariant knowledge. Then, in the second stage, we further enhance the model for better generalization on the unlabeled target set. This is achieved by exploring Intra-domain Point-MixUp in semi-supervised learning, which essentially regularizes the pseudo label distribution. Experiments from Waymo to nuScenes show that, with only 10% labeled target data, our SSDA3D can surpass the fully-supervised oracle model with 100% target label. Our code is available at https://github.com/yinjunbo/SSDA3D.

As acquiring manual labels on data could be costly, unsupervised domain adaptation (UDA), which transfers knowledge learned from a rich-label dataset to the unlabeled target dataset, is gaining increasingly more popularity. While extensive studies have been devoted to improving the model accuracy on target domain, an important issue of model robustness is neglected. To make things worse, conventional adversarial training (AT) methods for improving model robustness are inapplicable under UDA scenario since they train models on adversarial examples that are generated by supervised loss function. In this paper, we present a new meta self-training pipeline, named SRoUDA, for improving adversarial robustness of UDA models. Based on self-training paradigm, SRoUDA starts with pre-training a source model by applying UDA baseline on source labeled data and taraget unlabeled data with a developed random masked augmentation (RMA), and then alternates between adversarial target model training on pseudo-labeled target data and fine-tuning source model by a meta step. While self-training allows the direct incorporation of AT in UDA, the meta step in SRoUDA further helps in mitigating error propagation from noisy pseudo labels. Extensive experiments on various benchmark datasets demonstrate the state-of-the-art performance of SRoUDA where it achieves significant model robustness improvement without harming clean accuracy.

Despite the advance in deep learning algorithms, implementing supervised learning algorithms in medical datasets is difficult owing to the medical data's properties. This paper proposes SR-AnoGAN, which could generate higher resolution images and conduct anomaly detection more efficiently than AnoGAN. The most distinctive part of the proposed model is incorporating CNN and SRGAN into AnoGAN for reconstructing high-resolution images. Experimental results from X-ray datasets(pneumonia, covid-19) verify that the SR-AnoGAN outperforms the previous AnoGAN model through qualitative and quantitative approaches. Therefore, this paper shows the possibility of resolving data imbalance problems prevalent in the medical field, and proposing more precise diagnosis.

Existing multimodal conversation agents have shown impressive abilities to locate absolute positions or retrieve attributes in simple scenarios, but they fail to perform well when complex relative positions and information alignments are involved, which poses a bottleneck in response quality. In this paper, we propose a Situated Conversation Agent Pretrained with Multimodal Questions from Incremental Layout Graph (SPRING) with abilities of reasoning multi-hops spatial relations and connecting them with visual attributes in crowded situated scenarios. Specifically, we design two types of Multimodal Question Answering (MQA) tasks to pretrain the agent. All QA pairs utilized during pretraining are generated from novel Increment Layout Graphs (ILG). QA pair difficulty labels automatically annotated by ILG are used to promote MQA-based Curriculum Learning. Experimental results verify the SPRING's effectiveness, showing that it significantly outperforms state-of-the-art approaches on both SIMMC 1.0 and SIMMC 2.0 datasets. We release our code and data at https://github.com/LYX0501/SPRING.

This demonstration introduces SOREO, a system that explores the possibility of extending UAVs autonomy through machine learning. It brings a contribution to the following problem: Having a fleet of drones and a geographic area, how to learn the shortest paths between any point with regards to the base points for optimal and safe package delivery? Starting from a set of possible actions, a virtual design of a geographic location of interest, e.g., a city, and a reward value, SOREO is capable of learning not only how to prevent collisions with obstacles, e.g., walls and buildings, but also to find the shortest path between any two points, i.e., the base and the target. SOREO exploits based on the Q-learning algorithm.

In the context of verification of data-aware processes, a formal approach based on satisfiability modulo theories (SMT) has been considered to verify parameterised safety properties. This approach requires a combination of model-theoretic notions and algorithmic techniques based on backward reachability. We introduce here Ontology-Based Processes, which are a variant of one of the most investigated models in this spectrum, namely simple artifact systems (SASs), where, instead of managing a database, we operate over a description logic (DL) ontology. We prove that when the DL is expressed in (a slight extension of) RDFS, it enjoys suitable model-theoretic properties, and that by relying on such DL we can define Ontology-Based Processes to which backward reachability can still be applied. Relying on these results we are able to show that in this novel setting, verification of safety properties is decidable in PSPACE.

Multivariate time series forecasting with hierarchical structure is widely used in real-world applications, e.g., sales predictions for the geographical hierarchy formed by cities, states, and countries. The hierarchical time series (HTS) forecasting includes two sub-tasks, i.e., forecasting and reconciliation. In the previous works, hierarchical information is only integrated in the reconciliation step to maintain coherency, but not in forecasting step for accuracy improvement. In this paper, we propose two novel tree-based feature integration mechanisms, i.e., top-down convolution and bottom-up attention to leverage the information of the hierarchical structure to improve the forecasting performance. Moreover, unlike most previous reconciliation methods which either rely on strong assumptions or focus on coherent constraints only, we utilize deep neural optimization networks, which not only achieve coherency without any assumptions, but also allow more flexible and realistic constraints to achieve task-based targets, e.g., lower under-estimation penalty and meaningful decision-making loss to facilitate the subsequent downstream tasks. Experiments on real-world datasets demonstrate that our tree-based feature integration mechanism achieves superior performances on hierarchical forecasting tasks compared to the state-of-the-art methods, and our neural optimization networks can be applied to real-world tasks effectively without any additional effort under coherence and task-based constraints.

Exploration into quantum machine learning has grown tremendously in recent years due to the ability of quantum computers to speed up classical programs. However, these ef- forts have yet to solve unsupervised similarity detection tasks due to the challenge of porting them to run on quantum com- puters. To overcome this challenge, we propose SLIQ, the first open-sourced work for resource-efficient quantum sim- ilarity detection networks, built with practical and effective quantum learning and variance-reducing algorithms.

Emotion recognition in conversation (ERC) has received increasing attention from the research community. However, the ERC task is challenging, largely due to the complex and unstructured properties of multi-party conversations. Besides, the majority of daily dialogues take place in a specific context or circumstance, which requires rich external knowledge to understand the background of a certain dialogue. In this paper, we address these challenges by explicitly modeling the discourse relations between utterances and incorporating symbolic knowledge into multi-party conversations. We first introduce a dialogue parsing algorithm into ERC and further improve the algorithm through a transfer learning method. Moreover, we leverage different symbolic knowledge graph relations to learn knowledge-enhanced features for the ERC task. Extensive experiments on three benchmarks demonstrate that both dialogue structure graphs and symbolic knowledge are beneficial to the model performance on the task. Additionally, experimental results indicate that the proposed model surpasses baseline models on several indices.

In this paper, we propose Stochastic Knowledge Distillation (SKD) to obtain compact BERT-style language model dubbed SKDBERT. In each distillation iteration, SKD samples a teacher model from a pre-defined teacher team, which consists of multiple teacher models with multi-level capacities, to transfer knowledge into student model in an one-to-one manner. Sampling distribution plays an important role in SKD. We heuristically present three types of sampling distributions to assign appropriate probabilities for multi-level teacher models. SKD has two advantages: 1) it can preserve the diversities of multi-level teacher models via stochastically sampling single teacher model in each distillation iteration, and 2) it can also improve the efficacy of knowledge distillation via multi-level teacher models when large capacity gap exists between the teacher model and the student model. Experimental results on GLUE benchmark show that SKDBERT reduces the size of a BERT model by 40% while retaining 99.5% performances of language understanding and being 100% faster.

We propose a novel solution for unpaired image-to-image (I2I) translation. To translate complex images with a wide range of objects to a different domain, recent approaches often use the object annotations to perform per-class source-to-target style mapping. However, there remains a point for us to exploit in the I2I. An object in each class consists of multiple components, and all the sub-object components have different characteristics. For example, a car in CAR class consists of a car body, tires, windows and head and tail lamps, etc., and they should be handled separately for realistic I2I translation. The simplest solution to the problem will be to use more detailed annotations with sub-object component annotations than the simple object annotations, but it is not possible. The key idea of this paper is to bypass the sub-object component annotations by leveraging the original style of the input image because the original style will include the information about the characteristics of the sub-object components. Specifically, for each pixel, we use not only the per-class style gap between the source and target domains but also the pixel’s original style to determine the target style of a pixel. To this end, we present Style Harmonization for unpaired I2I translation (SHUNIT). Our SHUNIT generates a new style by harmonizing the target domain style retrieved from a class memory and an original source image style. Instead of direct source-to-target style mapping, we aim for source and target styles harmonization. We validate our method with extensive experiments and achieve state-of-the-art performance on the latest benchmark sets. The source code is available online: https://github.com/bluejangbaljang/SHUNIT.