查看更多>>摘要：With the explosion of sensory data in the Internet of Things(IoT)era,conventional machine vision systems are becoming increasingly difficult to meet the requirements of high efficiency,low energy consumption,and low latency due to their inherent shortcomings of separate sensing,memory,and computing units.Inspired by the retina and neuromorphic computing,the sensing-memory-computing integrated vision system(SMCVS)that features low power consumption,low latency,and high parallelism has been considered a promising technology to surpass the von Neumann architecture and realize strong artificial intelligence.Meanwhile,novel materials like two-dimensional semiconductors and quantum dots with novel optoelectronic performance provide hardware carriers for implementing integrated sensing-memory-computing architectures,attracting considerable attention.This paper reviews the recent research progress in bioinspired vision systems in terms of biomimetic mechanisms,design principles,computational architectures,and applications.Firstly,the biomimetic mechanisms are illustrated to guide the design of high-performance artificial visual perception systems.Then the research progress of optoelectronic-synapse-based bioinspired vision systems in the device principles and applications including image filtering,color recognition,visual adaptation,and motion detection are summarized.Finally,the challenges and future developing directions of the SMCVS are provided regarding bionic application,architecture design,and device fabrication.
查看更多>>摘要：Domain generalization aims to learn a model that is generalizable to an unseen target domain,which is a fundamental and challenging task in machine learning for out-of-distribution generalization.This paper proposes a novel domain generalization approach that enforces the learned model to be able to gen-eralize well over the train/val subset splitting of the training dataset.This idea is modeled herein as an adversarial data splitting framework,formulated as a min-max optimization problem inspired by the meta-learning approach.The min-max optimization problem is solved by iteratively splitting the training dataset into the training and val subsets to maximize the domain shift measured by the objective function and updating the model parameters to enable the model to generalize well from the training subset to the val subset by minimizing the objective function.This adversarial training approach does not assume the known domain labels of the training data;instead,it automatically investigates the"hard"splitting of the train/val subsets to learn the generalizable model.Extensive experimental results using three benchmark datasets demonstrate the superiority of this approach.In addition,we derive a generalization error bound for the theoretical understanding of our proposed approach.
查看更多>>摘要：In this paper,we take the advantage of previous pre-trained models(PTMs)and propose a novel Chinese pre-trained unbalanced transformer(CPT).Different from previous Chinese PTMs,CPT is designed to utilize the shared knowledge between natural language understanding(NLU)and natural language generation(NLG)to boost the performance.CPT consists of three parts:a shared encoder,an understanding decoder,and a generation decoder.Two specific decoders with a shared encoder are pre-trained with masked language modeling(MLM)and denoising auto-encoding(DAE)tasks,respectively.With the partially shared architecture and multi-task pre-training,CPT can(1)learn specific knowledge of both NLU or NLG tasks with two decoders and(2)be fine-tuned flexibly that fully exploits the potential of the model.Moreover,the unbalanced transformer saves the computational and storage cost,which makes CPT competitive and greatly accelerates the inference of text generation.Experimental results on a wide range of Chinese NLU and NLG tasks show the effectiveness of CPT.
查看更多>>摘要：Discovering causality from multivariate time series is an important but challenging problem.Most existing methods focus on estimating the Granger causal structures among multivariate time series,while ignoring the prior knowledge of these time series,e.g.,the group of the time series.Focusing on discov-ering the Granger causal structures among groups of time series,we propose a Granger causal representation learning method to solve this problem.First,we use the multiset canonical correlation analysis method to learn the Granger causal representation of each group of time series.Then,we model the Granger causal relationships among the learned Granger causal representations using a recurrent neural network with tem-poral information.Finally,we formulate the above two stages into one unified optimization problem,which is efficiently solved using the augmented Lagrangian method.We conduct extensive experiments on synthetic and real-world datasets to validate the correctness and effectiveness of the proposed method.
查看更多>>摘要：Deep reinforcement learning models are vulnerable to adversarial attacks that can decrease the cumulative expected reward of a victim by manipulating its observations.Despite the efficiency of previous optimization-based methods for generating adversarial noise in supervised learning,such methods might not achieve the lowest cumulative reward since they do not generally explore the environmental dynamics.Herein,a framework is provided to better understand the existing methods by reformulating the problem of adversarial attacks on reinforcement learning in the function space.The reformulation approach adopted herein generates an optimal adversary in the function space of targeted attacks,repelling them via a generic two-stage framework.In the first stage,a deceptive policy is trained by hacking the environment and discovering a set of trajectories routing to the lowest reward or the worst-case performance.Next,the adversary misleads the victim to imitate the deceptive policy by perturbing the observations.Compared to existing approaches,it is theoretically shown that our adversary is strong under an appropriate noise level.Extensive experiments demonstrate the superiority of the proposed method in terms of efficiency and effectiveness,achieving state-of-the-art performance in both Atari and MuJoCo environments.
查看更多>>摘要：Span-based joint extraction simultaneously conducts named entity recognition(NER)and re-lation extraction(RE)in a text span form.However,since previous span-based models rely on span-level classifications,they cannot benefit from token-level label information,which has been proven advantageous for the task.In this paper,we propose a sequence tagging augmented span-based network(STSN),a span-based joint model that can make use of token-level label information.In STSN,we construct a core neural architecture by deep stacking multiple attention layers,each of which consists of three basic attention units.On the one hand,the core architecture enables our model to learn token-level label information via the sequence tagging mechanism and then uses the information in the span-based joint extraction;on the other hand,it establishes a bi-directional information interaction between NER and RE.Experimental results on three benchmark datasets show that STSN consistently outperforms the strongest baselines in terms of F1,creating new state-of-the-art results.
查看更多>>摘要：Intelligent education is a significant application of artificial intelligence.One of the key research topics in intelligence education is cognitive diagnosis,which aims to gauge the level of proficiency among students on specific knowledge concepts(e.g.,Geometry).To the best of our knowledge,most of the existing cognitive models primarily focus on improving diagnostic accuracy while rarely considering fairness issues;for instance,the diagnosis of students may be affected by various sensitive attributes(e.g.,region).In this paper,we aim to explore fairness in cognitive diagnosis and answer two questions:(1)Are the results of existing cognitive diagnosis models affected by sensitive attributes?(2)If yes,how can we mitigate the impact of sensitive attributes to ensure fair diagnosis results?To this end,we first empirically reveal that several well-known cognitive diagnosis methods usually lead to unfair performances,and the trend of unfairness varies among different cognitive diagnosis models.Then,we make a theoretical analysis to explain the reasons behind this phenomenon.To resolve the unfairness problem in existing cognitive diagnosis models,we propose a general fairness-aware cognitive diagnosis framework,FairCD.Our fundamental principle involves eliminating the effect of sensitive attributes on student proficiency.To achieve this,we divide student proficiency in existing cognitive diagnosis models into two components:bias proficiency and fair proficiency.We design two orthogonal tasks for each of them to ensure that fairness in proficiency remains independent of sensitive attributes and take it as the final diagnosed result.Extensive experiments on the Program for International Student Assessment(PISA)dataset clearly show the effectiveness of our framework.
查看更多>>摘要：This paper proposes an imitation learning method to learn a universal agent policy for unlabeled multi-agent pathfinding(unlabeled MAPF)in grid environments.The method transforms the unlabeled MAPF problem into a series of temporal-independent homogeneous classification problems for each agent.Based on this transformation,a neural network is designed to imitate a distance-optimal expert algorithm.The neural network consists of two successive modules:perception field learner and field integrating classifier.The former refines and encodes the current system state into a perception field for each agent by combining a set of learnable field-generating functions.The latter takes an agent's perception field as input and decides the agent's next action based on a triplet cross-attention mechanism.We evaluate our method on a diverse set of unlabeled MAPF tasks.Compared with state-of-the-art counterparts,the experimental results manifest the superiority of the proposed method in both generalization ability and scalability.
查看更多>>摘要：Unsupervised domain adaptation(UDA)studies how to transfer a learner from a labeled source domain to an unlabeled target domain with different distributions.Existing methods mainly focus on match-ing marginal distributions of the source and target domains,which probably leads to a misalignment of sam-ples from the same class but different domains.In this paper,we tackle this misalignment issue by achieving the class-conditioned transferring from a new perspective.Specifically,we propose a method named maximiz-ing conditional independence(MCI)for UDA,which maximizes the conditional independence of feature and domain given class in the reproducing kernel Hilbert spaces.The optimization of conditional independence can be viewed as a surrogate for minimizing class-wise mutual information between feature and domain.An interpretable empirical estimation of the conditional dependence measure is deduced and connected with the unconditional case.Besides,we provide an upper bound on the target error by taking the class-conditional distribution into account,which provides a new theoretical insight for class-conditioned transferring.Ex-tensive experiments on six benchmark datasets and various ablation studies validate the effectiveness of the proposed model in dealing with UDA.
查看更多>>摘要：At present,one typical control strategy for guaranteeing transient and steady-state performance is funnel control and prescribed performance control.The strategy features completely discarding the sys-tem nonlinearities,even if they are completely known and available.Such an intrinsic feature requires the controller to produce a larger control effect to eliminate the negative impact caused by the high nonlinear-ities,leading to a conservative controller.In this paper,we fully take advantage of known information on nonlinearities in control design,instead of completely discarding it as done in funnel control.Particularly,we leverage adaptive techniques(i.e.,high-gain dynamic compensation)to deal with unknown system nonlinear-ities.Meanwhile,we integrate the tools of output feedback,tracking control,and the performance guarantee.As such,an adaptive output-feedback scheme is developed for global tracking with spatiotemporal perfor-mance specifications:arbitrarily given tracking accuracy and accuracy-ensured time.A simulation example supports the developed approach.
NETL
NSTL
万方数据