AAAI 2023 论文列表

We study risk-sensitive reinforcement learning (RL) based on an entropic risk measure in episodic non-stationary Markov decision processes (MDPs). Both the reward functions and the state transition kernels are unknown and allowed to vary arbitrarily over time with a budget on their cumulative variations. When this variation budget is known a prior, we propose two restart-based algorithms, namely Restart-RSMB and Restart-RSQ, and establish their dynamic regrets. Based on these results, we further present a meta-algorithm that does not require any prior knowledge of the variation budget and can adaptively detect the non-stationarity on the exponential value functions. A dynamic regret lower bound is then established for non-stationary risk-sensitive RL to certify the near-optimality of the proposed algorithms. Our results also show that the risk control and the handling of the non-stationarity can be separately designed in the algorithm if the variation budget is known a prior, while the non-stationary detection mechanism in the adaptive algorithm depends on the risk parameter. This work offers the first non-asymptotic theoretical analyses for the non-stationary risk-sensitive RL in the literature.

Parallel tempering (PT), also known as replica exchange, is the go-to workhorse for simulations of multi-modal distributions. The key to the success of PT is to adopt efficient swap schemes. The popular deterministic even-odd (DEO) scheme exploits the non-reversibility property and has successfully reduced the communication cost from quadratic to linear given the sufficiently many chains. However, such an innovation largely disappears in big data due to the limited chains and few bias-corrected swaps. To handle this issue, we generalize the DEO scheme to promote non-reversibility and propose a few solutions to tackle the underlying bias caused by the geometric stopping time. Notably, in big data scenarios, we obtain a nearly linear communication cost based on the optimal window size. In addition, we also adopt stochastic gradient descent (SGD) with large and constant learning rates as exploration kernels. Such a user-friendly nature enables us to conduct approximation tasks for complex posteriors without much tuning costs.

Reinforcement learning methods typically discount future rewards using an exponential scheme to achieve theoretical convergence guarantees. Studies from neuroscience, psychology, and economics suggest that human and animal behavior is better captured by the hyperbolic discounting model. Hyperbolic discounting has recently been studied in deep reinforcement learning and has shown promising results. However, this area of research is seemingly understudied, with most extant and continuing research using the standard exponential discounting formulation. My dissertation examines the effects of non-exponential discounting functions (such as hyperbolic) on an agent's learning and aims to investigate their impact on multi-agent systems and generalization tasks. A key objective of this study is to link the discounting rate to an agent's approximation of the underlying hazard rate of its environment through survival analysis.

Transfer learning refers to the transfer of knowledge or information from a relevant source domain to a target domain. However, most existing transfer learning theories and algorithms focus on IID tasks, where the source/target samples are assumed to be independent and identically distributed. Very little effort is devoted to theoretically studying the knowledge transferability on non-IID tasks, e.g., cross-network mining. To bridge the gap, in this paper, we propose rigorous generalization bounds and algorithms for cross-network transfer learning from a source graph to a target graph. The crucial idea is to characterize the cross-network knowledge transferability from the perspective of the Weisfeiler-Lehman graph isomorphism test. To this end, we propose a novel Graph Subtree Discrepancy to measure the graph distribution shift between source and target graphs. Then the generalization error bounds on cross-network transfer learning, including both cross-network node classification and link prediction tasks, can be derived in terms of the source knowledge and the Graph Subtree Discrepancy across domains. This thereby motivates us to propose a generic graph adaptive network (GRADE) to minimize the distribution shift between source and target graphs for cross-network transfer learning. Experimental results verify the effectiveness and efficiency of our GRADE framework on both cross-network node classification and cross-domain recommendation tasks.

Deepfake brings huge and potential negative impacts to our daily lives. As the real-life Deepfake videos circulated on the Internet become more authentic, most existing detection algorithms have failed since few visual differences can be observed between an authentic video and a Deepfake one. However, the forensic traces are always retained within the synthesized videos. In this study, we present a noise-based Deepfake detection model, NoiseDF for short, which focuses on the underlying forensic noise traces left behind the Deepfake videos. In particular, we enhance the RIDNet denoiser to extract noise traces and features from the cropped face and background squares of the video image frames. Meanwhile, we devise a novel Multi-Head Relative-Interaction method to evaluate the degree of interaction between the faces and backgrounds that plays a pivotal role in the Deepfake detection task. Besides outperforming the state-of-the-art models, the visualization of the extracted Deepfake forensic noise traces has further displayed the evidence and proved the robustness of our approach.

Next Point-Of-Interest (POI) recommendation plays an important role in various location-based services. Its main objective is to predict the user's next interested POI based on her previous check-in information. Most existing methods directly use users' historical check-in trajectories to construct various graphs to assist sequential models to complete this task. However, as users' check-in data is extremely sparse, it is difficult to capture the potential relations between POIs by directly using these check-in data. To this end, we propose the Sequence-based Neighbour search and Prediction Model (SNPM) for next POI recommendation. In SNPM, the RotatE knowledge graph embedding and Eigenmap methods are used to extract POI relationships implied in check-in data, and build the POI similarity graph. Then, we enhance the model's generalized representations of POIs' general features by aggregating similar POIs. As the context is typically rich and valuable when making Next POI predictions, the sequence model selects which POIs to aggregate not only depends on the current state, but also needs to consider the previous POI sequence. Therefore, we construct a Sequence-based, Dynamic Neighbor Graph (SDNG) to find the similarity neighbourhood and develop a Multi-Step Dependency Prediction model (MSDP) inspired by RotatE, which explicitly leverage information from previous states. We evaluate the proposed model on two real-world datasets, and the experimental results show that the proposed method significantly outperforms existing state-of-the-art POI recommendation methods.

Non-governmental organizations for environmental conservation have a significant interest in monitoring conservation-related media and getting timely updates about infrastructure construction projects as they may cause massive impact to key conservation areas. Such monitoring, however, is difficult and time-consuming. We introduce NewsPanda, a toolkit which automatically detects and analyzes online articles related to environmental conservation and infrastructure construction. We fine-tune a BERT-based model using active learning methods and noise correction algorithms to identify articles that are relevant to conservation and infrastructure construction. For the identified articles, we perform further analysis, extracting keywords and finding potentially related sources. NewsPanda has been successfully deployed by the World Wide Fund for Nature teams in the UK, India, and Nepal since February 2022. It currently monitors over 80,000 websites and 1,074 conservation sites across India and Nepal, saving more than 30 hours of human efforts weekly. We have now scaled it up to cover 60,000 conservation sites globally.

Knowledge representation and reasoning in neural networks has been a long-standing endeavour which has attracted much attention recently. The principled integration of reasoning and learning in neural networks is a main objective of the area of neurosymbolic Artificial Intelligence. In this paper, a neurosymbolic system is introduced that can represent any propositional logic formula. A proof of equivalence is presented showing that energy minimization in restricted Boltzmann machines corresponds to logical reasoning. We demonstrate the application of our approach empirically on logical reasoning and learning from data and knowledge. Experimental results show that reasoning can be performed effectively for a class of logical formulae. Learning from data and knowledge is also evaluated in comparison with learning of logic programs using neural networks. The results show that our approach can improve on state-of-the-art neurosymbolic systems. The theorems and empirical results presented in this paper are expected to reignite the research on the use of neural networks as massively-parallel models for logical reasoning and promote the principled integration of reasoning and learning in deep networks.

Travelling salesman problem (TSP) is NP-Hard with exponential search space. Recently, the adoption of encoder-decoder models as neural TSP solvers has emerged as an attractive topic because they can instantly obtain near-optimal results for small-scale instances. Nevertheless, their training efficiency and solution quality degrade dramatically when dealing with large-scale problems. To address the issue, we propose a novel progressive distillation framework, by adopting curriculum learning to train TSP samples in increasing order of their problem size and progressively distilling high-level knowledge from small models to large models via a distillation loss. In other words, the trained small models are used as the teacher network to guide action selection when training large models. To accelerate training speed, we also propose a Delaunary-graph based action mask and a new attention-based decoder to reduce decoding cost. Experimental results show that our approach establishes clear advantages over existing encoder-decoder models in terms of training effectiveness and solution quality. In addition, we validate its usefulness as an initial solution generator for the state-of-the-art TSP solvers, whose probability of obtaining the optimal solution can be further improved in such a hybrid manner.

Accurate estimation of output quantiles is crucial in many use cases, where it is desired to model the range of possibility. Modeling target distribution at arbitrary quantile levels and at arbitrary input attribute levels are important to offer a comprehensive picture of the data, and requires the quantile function to be expressive enough. The quantile function describing the target distribution using quantile levels is critical for quantile regression. Although various parametric forms for the distributions (that the quantile function specifies) can be adopted, an everlasting problem is selecting the most appropriate one that can properly approximate the data distributions. In this paper, we propose a non-parametric and data-driven approach, Neural Spline Search (NSS), to represent the observed data distribution without parametric assumptions. NSS is flexible and expressive for modeling data distributions by transforming the inputs with a series of monotonic spline regressions guided by symbolic operators. We demonstrate that NSS outperforms previous methods on synthetic, real-world regression and time-series forecasting tasks.

We leverage probabilistic models of neural representations to investigate how residual networks fit classes. To this end, we estimate class-conditional density models for representations learned by deep ResNets. We then use these models to characterize distributions of representations across learned classes. Surprisingly, we find that classes in the investigated models are not fitted in a uniform way. On the contrary: we uncover two groups of classes that are fitted with markedly different distributions of representations. These distinct modes of class-fitting are evident only in the deeper layers of the investigated models, indicating that they are not related to low-level image features. We show that the uncovered structure in neural representations correlate with memorization of training examples and adversarial robustness. Finally, we compare class-conditional distributions of neural representations between memorized and typical examples. This allows us to uncover where in the network structure class labels arise for memorized and standard inputs.

Neural networks (NN) are an increasingly important representation of action policies pi. Recent work has extended predicate abstraction to prove safety of such pi, through policy predicate abstraction (PPA) which over-approximates the state space subgraph induced by pi. The advantage of PPA is that reasoning about the NN – calls to SMT solvers – is required only locally, at individual abstract state transitions, in contrast to bounded model checking (BMC) where SMT must reason globally about sequences of NN decisions. Indeed, it has been shown that PPA can outperform a simple BMC implementation. However, the abstractions underlying these results (i.e., the abstraction predicates) were supplied manually. Here we automate this step. We extend counterexample guided abstraction refinement (CEGAR) to PPA. This involves dealing with a new source of spuriousness in abstract unsafe paths, pertaining not to transition behavior but to the decisions of the neural network pi. We introduce two methods tackling this issue based on the states involved, and we show that global SMT calls deciding spuriousness exactly can be avoided. We devise algorithmic enhancements leveraging incremental computation and heuristic search. We show empirically that the resulting verification tool has significant advantages over an encoding into the state-of-the-art model checker nuXmv. In particular, ours is the only approach in our experiments that succeeds in proving policies safe.

The query-document term matching plays an important role in information retrieval. However, the retrieval performance degrades when the documents get matched with the extraneous terms of the query which frequently arises in verbose queries. To address this problem, we generate the dense vector of the entire query and individual query terms using the pre-trained BERT (Bidirectional Encoder Representations from Transformers) model and subsequently analyze their relation to focus on the central terms. We then propose a context-aware attentive extension of unsupervised Markov Random Field-based sequential term dependence model that explicitly pays more attention to those contextually central terms. The proposed model utilizes the strengths of the pre-trained large language model for estimating the attention weight of terms and rank the documents in a single pass without any supervision.

Modeling continuous dynamical systems from discretely sampled observations is a fundamental problem in data science. Often, such dynamics are the result of non-local processes that present an integral over time. As such, these systems are modeled with Integro-Differential Equations (IDEs); generalizations of differential equations that comprise both an integral and a differential component. For example, brain dynamics are not accurately modeled by differential equations since their behavior is non-Markovian, i.e. dynamics are in part dictated by history. Here, we introduce the Neural IDE (NIDE), a novel deep learning framework based on the theory of IDEs where integral operators are learned using neural networks. We test NIDE on several toy and brain activity datasets and demonstrate that NIDE outperforms other models. These tasks include time extrapolation as well as predicting dynamics from unseen initial conditions, which we test on whole-cortex activity recordings in freely behaving mice. Further, we show that NIDE can decompose dynamics into their Markovian and non-Markovian constituents, via the learned integral operator, which we test on fMRI brain activity recordings of people on ketamine. Finally, the integrand of the integral operator provides a latent space that gives insight into the underlying dynamics, which we demonstrate on wide-field brain imaging recordings. Altogether, NIDE is a novel approach that enables modeling of complex non-local dynamics with neural networks.

Representing 3D objects and scenes with neural radiance fields has become very popular over the last years. Recently, surface-based representations have been proposed, that allow to reconstruct 3D objects from simple photographs. However, most current techniques require an accurate camera calibration, i.e. camera parameters corresponding to each image, which is often a difficult task to do in real-life situations. To this end, we propose a method for learning 3D surfaces from noisy camera parameters. We show that we can learn camera parameters together with learning the surface representation, and demonstrate good quality 3D surface reconstruction even with noisy camera observations.

Topic models and all their variants analyse text by learning meaningful representations through word co-occurrences. As pointed out by previous work, such models implicitly assume that the probability of a topic to be active and its proportion within each document are positively correlated. This correlation can be strongly detrimental in the case of documents created over time, simply because recent documents are likely better described by new and hence rare topics. In this work we leverage recent advances in neural variational inference and present an alternative neural approach to the dynamic Focused Topic Model. Indeed, we develop a neural model for topic evolution which exploits sequences of Bernoulli random variables in order to track the appearances of topics, thereby decoupling their activities from their proportions. We evaluate our model on three different datasets (the UN general debates, the collection of NeurIPS papers, and the ACL Anthology dataset) and show that it (i) outperforms state-of-the-art topic models in generalization tasks and (ii) performs comparably to them on prediction tasks, while employing roughly the same number of parameters, and converging about two times faster.

Normalizing flows have been successfully modeling a complex probability distribution as an invertible transformation of a simple base distribution. However, there are often applications that require more than invertibility. For instance, the computation of energies and forces in physics requires the second derivatives of the transformation to be well-defined and continuous. Smooth normalizing flows employ infinitely differentiable transformation, but with the price of slow non-analytic inverse transforms. In this work, we propose diffeomorphic non-uniform B-spline flows that are at least twice continuously differentiable while bi-Lipschitz continuous, enabling efficient parametrization while retaining analytic inverse transforms based on a sufficient condition for diffeomorphism. Firstly, we investigate the sufficient condition for C(k-2)-diffeomorphic non-uniform kth-order B-spline transformations. Then, we derive an analytic inverse transformation of the non-uniform cubic B-spline transformation for neural diffeomorphic non-uniform B-spline flows. Lastly, we performed experiments on solving the force matching problem in Boltzmann generators, demonstrating that our C2-diffeomorphic non-uniform B-spline flows yielded solutions better than previous spline flows and faster than smooth normalizing flows. Our source code is publicly available at https://github.com/smhongok/Non-uniform-B-spline-Flow.

One major limitation of CNNs is that they are vulnerable to adversarial attacks. Currently, adversarial robustness in neural networks is commonly optimized with respect to a small pre-selected adversarial noise strength, causing them to have potentially limited performance when under attack by larger adversarial noises in real-world scenarios. In this research, we aim to find Neural Architectures that have improved robustness on a wide range of adversarial noise strengths through Neural Architecture Search. In detail, we propose a lightweight Adversarial Noise Estimator to reduce the high cost of generating adversarial noise with respect to different strengths. Besides, we construct an Efficient Wide Spectrum Searcher to reduce the cost of adjusting network architecture with the large adversarial validation set during the search. With the two components proposed, the number of adversarial noise strengths searched can be increased significantly while having a limited increase in search time. Extensive experiments on benchmark datasets such as CIFAR and ImageNet demonstrate that with a significantly richer search signal in robustness, our method can find architectures with improved overall robustness while having a limited impact on natural accuracy and around 40% reduction in search time compared with the naive approach of searching. Codes available at: https://github.com/zhicheng2T0/Wsr-NAS.git

Restless multi-armed bandits are often used to model budget-constrained resource allocation tasks where receipt of the resource is associated with an increased probability of a favorable state transition. Prior work assumes that individual arms only benefit if they receive the resource directly. However, many allocation tasks occur within communities and can be characterized by positive externalities that allow arms to derive partial benefit when their neighbor(s) receive the resource. We thus introduce networked restless bandits, a novel multi-armed bandit setting in which arms are both restless and embedded within a directed graph. We then present Greta, a graph-aware, Whittle index-based heuristic algorithm that can be used to efficiently construct a constrained reward-maximizing action vector at each timestep. Our empirical results demonstrate that Greta outperforms comparison policies across a range of hyperparameter values and graph topologies. Code and appendices are available at https://github.com/crherlihy/networked_restless_bandits.