AAAI 2023 Papers

Partial Label (PL) learning refers to the task of learning from the partially labeled data, where each training instance is ambiguously equipped with a set of candidate labels but only one is valid. Advances in the recent deep PL learning literature have shown that the deep learning paradigms, e.g., self-training, contrastive learning, or class activate values, can achieve promising performance. Inspired by the impressive success of deep Semi-Supervised (SS) learning, we transform the PL learning problem into the SS learning problem, and propose a novel PL learning method, namely Partial Label learning with Semi-supervised Perspective (PLSP). Specifically, we first form the pseudo-labeled dataset by selecting a small number of reliable pseudo-labeled instances with high-confidence prediction scores and treating the remaining instances as pseudo-unlabeled ones. Then we design a SS learning objective, consisting of a supervised loss for pseudo-labeled instances and a semantic consistency regularization for pseudo-unlabeled instances. We further introduce a complementary regularization for those non-candidate labels to constrain the model predictions on them to be as small as possible. Empirical results demonstrate that PLSP significantly outperforms the existing PL baseline methods, especially on high ambiguity levels. Code available: https://github.com/changchunli/PLSP.

As it is cumbersome and expensive to acquire a huge amount of data for training neural dialog models, data augmentation is proposed to effectively utilize existing training samples. However, current data augmentation techniques on the dialog generation task mostly augment all cases in the training dataset without considering the intrinsic attributes between different cases. We argue that not all cases are beneficial for augmentation task, and the cases suitable for augmentation should obey the following two attributes: (1) low-quality (the dialog model cannot generate a high-quality response for the case), (2) representative (the case should represent the property of the whole dataset). Herein, we explore this idea by proposing a Selective Data Augmentation framework (SDA) for the response generation task. SDA employs a dual adversarial network to select the lowest quality and most representative data points for augmentation in one stage. Extensive experiments conducted on two publicly available datasets, i.e., DailyDialog and OpenSubtitles, show that our framework can improve the response generation performance with respect to various metrics

Spike camera is a kind of neuromorphic sensor that uses a novel ``integrate-and-fire'' mechanism to generate a continuous spike stream to record the dynamic light intensity at extremely high temporal resolution. However, as a trade-off for high temporal resolution, its spatial resolution is limited, resulting in inferior reconstruction details. To address this issue, this paper develops a network (SpikeSR-Net) to super-resolve a high-resolution image sequence from the low-resolution binary spike streams. SpikeSR-Net is designed based on the observation model of spike camera and exploits both the merits of model-based and learning-based methods. To deal with the limited representation capacity of binary data, a pixel-adaptive spike encoder is proposed to convert spikes to latent representation to infer clues on intensity and motion. Then, a motion-aligned super resolver is employed to exploit long-term correlation, so that the dense sampling in temporal domain can be exploited to enhance the spatial resolution without introducing motion blur. Experimental results show that SpikeSR-Net is promising in super-resolving higher-quality images for spike camera.

Reward shaping (RS) is a powerful method in reinforcement learning (RL) for overcoming the problem of sparse or uninformative rewards. However, RS typically relies on manually engineered shaping-reward functions whose construc- tion is time-consuming and error-prone. It also requires domain knowledge which runs contrary to the goal of autonomous learning. We introduce Reinforcement Learning Optimising Shaping Algorithm (ROSA), an automated reward shaping framework in which the shaping-reward function is constructed in a Markov game between two agents. A reward-shaping agent (Shaper) uses switching controls to determine which states to add shaping rewards for more efficient learning while the other agent (Controller) learns the optimal policy for the task using these shaped rewards. We prove that ROSA, which adopts existing RL algorithms, learns to construct a shaping-reward function that is beneficial to the task thus ensuring efficient convergence to high performance policies. We demonstrate ROSA’s properties in three didactic experiments and show its superior performance against state-of-the-art RS algorithms in challenging sparse reward environments.

Many NLP tasks can be regarded as a selection problem from a set of options, such as classification tasks, multi-choice question answering, etc. Textual entailment (TE) has been shown as the state-of-the-art (SOTA) approach to dealing with those selection problems. TE treats input texts as premises (P), options as hypotheses (H), then handles the selection problem by modeling (P, H) pairwise. Two limitations: first, the pairwise modeling is unaware of other options, which is less intuitive since humans often determine the best options by comparing competing candidates; second, the inference process of pairwise TE is time-consuming, especially when the option space is large. To deal with the two issues, this work first proposes a contextualized TE model (Context-TE) by appending other k options as the context of the current (P, H) modeling. Context-TE is able to learn more reliable decision for the H since it considers various context. Second, we speed up Context-TE by coming up with Parallel-TE, which learns the decisions of multiple options simultaneously. Parallel-TE significantly improves the inference speed while keeping comparable performance with Context-TE. Our methods are evaluated on three tasks (ultra-fine entity typing, intent detection and multi-choice QA) that are typical selection problems with different sizes of options. Experiments show our models set new SOTA performance; particularly, Parallel-TE is faster than the pairwise TE by k times in inference.

Prototype-based interpretability methods provide intuitive explanations of model prediction by comparing samples to a reference set of memorized exemplars or typical representatives in terms of similarity. In the field of sequential data modeling, similarity calculations of prototypes are usually based on encoded representation vectors. However, due to highly recursive functions, there is usually a non-negligible disparity between the prototype-based explanations and the original input. In this work, we propose a Self-Explaining Selective Model (SESM) that uses a linear combination of prototypical concepts to explain its own predictions. The model employs the idea of case-based reasoning by selecting sub-sequences of the input that mostly activate different concepts as prototypical parts, which users can compare to sub-sequences selected from different example inputs to understand model decisions. For better interpretability, we design multiple constraints including diversity, stability, and locality as training objectives. Extensive experiments in different domains demonstrate that our method exhibits promising interpretability and competitive accuracy.

Sample re-weighting strategies provide a promising mechanism to deal with imperfect training data in machine learning, such as noisily labeled or class-imbalanced data. One such strategy involves formulating a bi-level optimization problem called the meta re-weighting problem, whose goal is to optimize performance on a small set of perfect pivotal samples, called meta samples. Many approaches have been proposed to efficiently solve this problem. However, all of them assume that a perfect meta sample set is already provided while we observe that the selections of meta sample set is performance-critical. In this paper, we study how to learn to identify such a meta sample set from a large, imperfect training set, that is subsequently cleaned and used to optimize performance in the meta re-weighting setting. We propose a learning framework which reduces the meta samples selection problem to a weighted K-means clustering problem through rigorously theoretical analysis. We propose two clustering methods within our learning framework, Representation-based clustering method (RBC) and Gradient-based clustering method (GBC), for balancing performance and computational efficiency. Empirical studies demonstrate the performance advantage of our methods over various baseline methods

This paper is part of the AAAI-23 New Faculty Highlights. In my presentation, I will introduce my research goal, which is to build machines that see, interact with, and reason about the physical world just like humans. This problem, which we call physical scene understanding, involves three key topics that bridge research in computer science, AI, robotics, cognitive science, and neuroscience: Perception, Physical Interaction, and Reasoning.

We study the problem of training a principal in a multi-agent general-sum game using reinforcement learning (RL). Learning a robust principal policy requires anticipating the worst possible strategic responses of other agents, which is generally NP-hard. However, we show that no-regret dynamics can identify these worst-case responses in poly-time in smooth games. We propose a framework that uses this policy evaluation method for efficiently learning a robust principal policy using RL. This framework can be extended to provide robustness to boundedly rational agents too. Our motivating application is automated mechanism design: we empirically demonstrate our framework learns robust mechanisms in both matrix games and complex spatiotemporal games. In particular, we learn a dynamic tax policy that improves the welfare of a simulated trade-and-barter economy by 15%, even when facing previously unseen boundedly rational RL taxpayers.

Maintaining engagement and consistency is particularly important in dialogue systems. Existing works have improved the performance of dialogue systems by intentionally learning interlocutor personas with sophisticated network structures. One issue with this approach is that it requires more personal corpora with annotations. Additionally, these models typically perform the next utterance prediction to generate a response but neglect the discourse coherence in the entire conversation. To address these issues, this study proposes a method of learning to memorize entailment and discourse relations for persona-consistent dialogue tasks. Entailment text pairs in natural language inference dataset were applied to learn latent entailment relations as external memories by premise-to-hypothesis generation task. Furthermore, an internal memory with a similar architecture was applied to the discourse information in the dialogue. Placing orthogonality restrictions on these two memory spaces ensures that the latent entailment relations remain dialogue-independent. Both memories collaborate to obtain entailment and discourse representation for the generation, allowing a deeper understanding of both consistency and coherence. Experiments on two large public datasets, PersonaChat and DSTC7-AVSD, demonstrated the effectiveness of the proposed method. Both automatic and human evaluations indicate that the proposed model outperforms several strong baselines in terms of both persona consistency and response coherence. Our source code is availabled at https://github.com/Chenrj233/LMEDR.

Recently, the joint learning framework (JOINT) integrates matching based transductive reasoning and online inductive learning to achieve accurate and robust semi-supervised video object segmentation (SVOS). However, using the mask embedding as the label to guide the generation of target features in the two branches may result in inadequate target representation and degrade the performance. Besides, how to reasonably fuse the target features in the two different branches rather than simply adding them together to avoid the adverse effect of one dominant branch has not been investigated. In this paper, we propose a novel framework that emphasizes Learning to Learn Better (LLB) target features for SVOS, termed LLB, where we design the discriminative label generation module (DLGM) and the adaptive fusion module to address these issues. Technically, the DLGM takes the background-filtered frame instead of the target mask as input and adopts a lightweight encoder to generate the target features, which serves as the label of the online few-shot learner and the value of the decoder in the transformer to guide the two branches to learn more discriminative target representation. The adaptive fusion module maintains a learnable gate for each branch, which reweighs the element-wise feature representation and allows an adaptive amount of target information in each branch flowing to the fused target feature, thus preventing one branch from being dominant and making the target feature more robust to distractor. Extensive experiments on public benchmarks show that our proposed LLB method achieves state-of-the-art performance.

A critical challenge for open-domain dialogue agents is to generate persona-relevant and consistent responses. Due to the nature of persona sparsity in conversation scenarios, previous persona-based dialogue agents trained with Maximum Likelihood Estimation tend to overlook the given personas and generate responses irrelevant or inconsistent with personas. To address this problem, we propose a two-stage coarse-to-fine persona-aware training framework to improve the persona consistency of a dialogue agent progressively. Specifically, our framework first trains the dialogue agent to answer the constructed persona-aware questions, making it highly sensitive to the personas to generate persona-relevant responses. Then the dialogue agent is further trained with a contrastive learning paradigm by explicitly perceiving the difference between the consistent and the generated inconsistent responses, forcing it to pay more attention to the key persona information to generate consistent responses. By applying our proposed training framework to several representative baseline models, experimental results show significant boosts on both automatic and human evaluation metrics, especially the consistency of generated responses.

In dialogue state tracking (DST), the exploitation of dialogue history is a crucial research direction, and the existing DST models can be divided into two categories: full-history models and partial-history models. Since the “select first, use later” mechanism explicitly filters the distracting information being passed to the downstream state prediction, the partial-history models have recently achieved a performance advantage over the full-history models. However, besides the redundant information, some critical dialogue context information was inevitably filtered out by the partial-history models simultaneously. To reconcile the contextual consideration with avoiding the introduction of redundant information, we propose DICE-DST, a model-agnostic module widely applicable to the partial-history DST models, which aims to strengthen the ability of context exploitation for the encoder of each DST model. Specifically, we first construct a teacher encoder and devise two contextual reasoning tasks to train it to acquire extensive dialogue contextual knowledge. Then we transfer the contextual knowledge from the teacher encoder to the student encoder via a novel turn-level attention-alignment distillation. Experimental results show that our approach extensively improves the performance of partial-history DST models and thereby achieves new state-of-the-art performance on multiple mainstream datasets while keeping high efficiency.

Scene Graph Generation (SGG) aims to capture the semantic information in an image and build a structured representation, which facilitates downstream tasks. The current challenge in SGG is to tackle the biased predictions caused by the long-tailed distribution of predicates. Since multiple predicates in SGG are coupled in an image, existing data re-balancing methods cannot completely balance the head and tail predicates. In this work, a decoupled learning framework is proposed for unbiased scene graph generation by using attribute-guided predicate features to construct a balanced training set. Specifically, the predicate recognition is decoupled into Predicate Feature Representation Learning (PFRL) and predicate classifier training with a class-balanced predicate feature set, which is constructed by our proposed Attribute-guided Predicate Feature Generation (A-PFG) model. In the A-PFG model, we first define the class labels of and corresponding visual feature as attributes to describe a predicate. Then the predicate feature and the attribute embedding are mapped into a shared hidden space by a dual Variational Auto-encoder (VAE), and finally the synthetic predicate features are forced to learn the contextual information in the attributes via cross reconstruction and distribution alignment. To demonstrate the effectiveness of our proposed method, our decoupled learning framework and A-PFG model are applied to various SGG models. The empirical results show that our method is substantially improved on all benchmarks and achieves new state-of-the-art performance for unbiased scene graph generation. Our code is available at https://github.com/wanglei0618/A-PFG.

Recent research suggests that combining AI models with a human expert can exceed the performance of either alone. The combination of their capabilities is often realized by learning to defer algorithms that enable the AI to learn to decide whether to make a prediction for a particular instance or defer it to the human expert. However, to accurately learn which instances should be deferred to the human expert, a large number of expert predictions that accurately reflect the expert's capabilities are required—in addition to the ground truth labels needed to train the AI. This requirement shared by many learning to defer algorithms hinders their adoption in scenarios where the responsible expert regularly changes or where acquiring a sufficient number of expert predictions is costly. In this paper, we propose a three-step approach to reduce the number of expert predictions required to train learning to defer algorithms. It encompasses (1) the training of an embedding model with ground truth labels to generate feature representations that serve as a basis for (2) the training of an expertise predictor model to approximate the expert's capabilities. (3) The expertise predictor generates artificial expert predictions for instances not yet labeled by the expert, which are required by the learning to defer algorithms. We evaluate our approach on two public datasets. One with "synthetically" generated human experts and another from the medical domain containing real-world radiologists' predictions. Our experiments show that the approach allows the training of various learning to defer algorithms with a minimal number of human expert predictions. Furthermore, we demonstrate that even a small number of expert predictions per class is sufficient for these algorithms to exceed the performance the AI and the human expert can achieve individually.

Subgraph isomorphism counting is an important problem on graphs, as many graph-based tasks exploit recurring subgraph patterns. Classical methods usually boil down to a backtracking framework that needs to navigate a huge search space with prohibitive computational cost. Some recent studies resort to graph neural networks (GNNs) to learn a low-dimensional representation for both the query and input graphs, in order to predict the number of subgraph isomorphisms on the input graph. However, typical GNNs employ a node-centric message passing scheme that receives and aggregates messages on nodes, which is inadequate in complex structure matching for isomorphism counting. Moreover, on an input graph, the space of possible query graphs is enormous, and different parts of the input graph will be triggered to match different queries. Thus, expecting a fixed representation of the input graph to match diversely structured query graphs is unrealistic. In this paper, we propose a novel GNN called Count-GNN for subgraph isomorphism counting, to deal with the above challenges. At the edge level, given that an edge is an atomic unit of encoding graph structures, we propose an edge-centric message passing scheme, where messages on edges are propagated and aggregated based on the edge adjacency to preserve fine-grained structural information. At the graph level, we modulate the input graph representation conditioned on the query, so that the input graph can be adapted to each query individually to improve their matching. Finally, we conduct extensive experiments on a number of benchmark datasets to demonstrate the superior performance of Count-GNN.

The ability to efficiently solve hard combinatorial optimization problems is a key prerequisite to various applications of declarative programming paradigms. Symmetries in solution candidates pose a significant challenge to modern optimization algorithms since the enumeration of such candidates might substantially reduce their performance. This paper proposes a novel approach using Inductive Logic Programming (ILP) to lift symmetry-breaking constraints for optimization problems modeled in Answer Set Programming (ASP). Given an ASP encoding with optimization statements and a set of small representative instances, our method augments ground ASP programs with auxiliary normal rules enabling the identification of symmetries using existing tools, like SBASS. Then, the obtained symmetries are lifted to first-order constraints with ILP. We prove the correctness of our method and evaluate it on real-world optimization problems from the domain of automated configuration. Our experiments show significant improvements of optimization performance due to the learned first-order constraints.

Successful machine learning methods require a trade-off between memorization and generalization. Too much memorization and the model cannot generalize to unobserved examples. Too much over-generalization and we risk under-fitting the data. While we commonly measure their performance through cross validation and accuracy metrics, how should these algorithms cope in domains that are extremely under-determined where accuracy is always unsatisfactory? We present a novel probabilistic graphical model structure learning approach that can learn, generalize and explain in these elusive domains by operating at the random variable instantiation level. Using Minimum Description Length (MDL) analysis, we propose a new decomposition of the learning problem over all training exemplars, fusing together minimal entropy inferences to construct a final knowledge base. By leveraging Bayesian Knowledge Bases (BKBs), a framework that operates at the instantiation level and inherently subsumes Bayesian Networks (BNs), we develop both a theoretical MDL score and associated structure learning algorithm that demonstrates significant improvements over learned BNs on 40 benchmark datasets. Further, our algorithm incorporates recent off-the-shelf DAG learning techniques enabling tractable results even on large problems. We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA).

Search engines are essential internet services, enabling users to efficiently find the information they need. Session search employs users’ session logs of queries to solve complex retrieval tasks, in which users search multiple times until interested documents are found. Most existing session search models focus on the contextual information within the current search, ignoring the evidence from historical search sessions. Considering the fact that many ongoing retrieval tasks should have already been carried out by other users with a similar intent, we argue that historical sessions with similar intents can help improve the accuracy of the current search task. We propose a novel Similar Session-enhanced Ranking (SSR) model to improve the session search performance using historical sessions with similar intents. Specifically, the candidate historical sessions are matched by query-level and session-level semantic similarity, and then query-level neighbor behaviors are aggregated by a Query-guided GNN (QGNN) while session-level neighbor behaviors are aggregated using the attention mechanism. Finally, we integrate the refined and aggregated historical neighbor information into the current search session. Experimental results on AOL and Tiangong-ST datasets show that our SSR model significantly outperforms the state-of-the-art models.