While multi-exit neural networks are regarded as a promising solution for making efficient inference via early exits, combating adversarial attacks remains a challenging problem. In multi-exit networks, due to the high dependency among different submodels, an adversarial example targeting a specific exit not only degrades the performance of the target exit but also reduces the performance of all other exits concurrently. This makes multi-exit networks highly vulnerable to simple adversarial attacks. In this paper, we propose NEO-KD, a knowledge-distillation-based adversarial training strategy that tackles this fundamental challenge based on two key contributions. NEO-KD first resorts to neighbor knowledge distillation to guide the output of the adversarial examples to tend to the ensemble outputs of neighbor exits of clean data. NEO-KD also employs exit-wise orthogonal knowledge distillation for reducing adversarial transferability across different submodels. The result is a significantly improved robustness against adversarial attacks. Experimental results on various datasets/models show that our method achieves the best adversarial accuracy with reduced computation budgets, compared to the baselines relying on existing adversarial training or knowledge distillation techniques for multi-exit networks.
相關內容
Lipschitz-constrained neural networks have several advantages over unconstrained ones and can be applied to a variety of problems, making them a topic of attention in the deep learning community. Unfortunately, it has been shown both theoretically and empirically that they perform poorly when equipped with ReLU activation functions. By contrast, neural networks with learnable 1-Lipschitz linear splines are known to be more expressive. In this paper, we show that such networks correspond to global optima of a constrained functional optimization problem that consists of the training of a neural network composed of 1-Lipschitz linear layers and 1-Lipschitz freeform activation functions with second-order total-variation regularization. Further, we propose an efficient method to train these neural networks. Our numerical experiments show that our trained networks compare favorably with existing 1-Lipschitz neural architectures.
Due to their ability to model meaningful higher order relations among a set of entities, higher order network models have emerged recently as a powerful alternative for graph-based network models which are only capable of modeling binary relationships. Message passing paradigm is still dominantly used to learn representations even for higher order network models. While powerful, message passing can have disadvantages during inference, particularly when the higher order connectivity information is missing or corrupted. To overcome such limitations, we propose Topo-MLP, a purely MLP-based simplicial neural network algorithm to learn the representation of elements in a simplicial complex without explicitly relying on message passing. Our framework utilizes a novel Higher Order Neighborhood Contrastive (HONC) loss which implicitly incorporates the simplicial structure into representation learning. Our proposed model's simplicity makes it faster during inference. Moreover, we show that our model is robust when faced with missing or corrupted connectivity structure.
Multi-agent and multi-robot systems (MRS) often rely on direct communication for information sharing. This work explores an alternative approach inspired by eavesdropping mechanisms in nature that involves casual observation of agent interactions to enhance decentralized knowledge dissemination. We achieve this through a novel IKT-BT framework tailored for a behavior-based MRS, encapsulating knowledge and control actions in Behavior Trees (BT). We present two new BT-based modalities - eavesdrop-update (EU) and eavesdrop-buffer-update (EBU) - incorporating unique eavesdropping strategies and efficient episodic memory management suited for resource-limited swarm robots. We theoretically analyze the IKT-BT framework for an MRS and validate the performance of the proposed modalities through extensive experiments simulating a search and rescue mission. Our results reveal improvements in both global mission performance outcomes and agent-level knowledge dissemination with a reduced need for direct communication.
Offline imitation learning (IL) refers to learning expert behavior solely from demonstrations, without any additional interaction with the environment. Despite significant advances in offline IL, existing techniques find it challenging to learn policies for long-horizon tasks and require significant re-training when task specifications change. Towards addressing these limitations, we present GO-DICE an offline IL technique for goal-conditioned long-horizon sequential tasks. GO-DICE discerns a hierarchy of sub-tasks from demonstrations and uses these to learn separate policies for sub-task transitions and action execution, respectively; this hierarchical policy learning facilitates long-horizon reasoning. Inspired by the expansive DICE-family of techniques, policy learning at both the levels transpires within the space of stationary distributions. Further, both policies are learnt with goal conditioning to minimize need for retraining when task goals change. Experimental results substantiate that GO-DICE outperforms recent baselines, as evidenced by a marked improvement in the completion rate of increasingly challenging pick-and-place Mujoco robotic tasks. GO-DICE is also capable of leveraging imperfect demonstration and partial task segmentation when available, both of which boost task performance relative to learning from expert demonstrations alone.
Efficiently capturing the complex spatiotemporal representations from large-scale unlabeled traffic data remains to be a challenging task. In considering of the dilemma, this work employs the advanced contrastive learning and proposes a novel Spatial-Temporal Synchronous Contextual Contrastive Learning (STS-CCL) model. First, we elaborate the basic and strong augmentation methods for spatiotemporal graph data, which not only perturb the data in terms of graph structure and temporal characteristics, but also employ a learning-based dynamic graph view generator for adaptive augmentation. Second, we introduce a Spatial-Temporal Synchronous Contrastive Module (STS-CM) to simultaneously capture the decent spatial-temporal dependencies and realize graph-level contrasting. To further discriminate node individuals in negative filtering, a Semantic Contextual Contrastive method is designed based on semantic features and spatial heterogeneity, achieving node-level contrastive learning along with negative filtering. Finally, we present a hard mutual-view contrastive training scheme and extend the classic contrastive loss to an integrated objective function, yielding better performance. Extensive experiments and evaluations demonstrate that building a predictor upon STS-CCL contrastive learning model gains superior performance than existing traffic forecasting benchmarks. The proposed STS-CCL is highly suitable for large datasets with only a few labeled data and other spatiotemporal tasks with data scarcity issue.
Compressing a predefined deep neural network (DNN) into a compact sub-network with competitive performance is crucial in the efficient machine learning realm. This topic spans various techniques, from structured pruning to neural architecture search, encompassing both pruning and erasing operators perspectives. Despite advancements, existing methods suffers from complex, multi-stage processes that demand substantial engineering and domain knowledge, limiting their broader applications. We introduce the third-generation Only-Train-Once (OTOv3), which first automatically trains and compresses a general DNN through pruning and erasing operations, creating a compact and competitive sub-network without the need of fine-tuning. OTOv3 simplifies and automates the training and compression process, minimizes the engineering efforts required from users. It offers key technological advancements: (i) automatic search space construction for general DNNs based on dependency graph analysis; (ii) Dual Half-Space Projected Gradient (DHSPG) and its enhanced version with hierarchical search (H2SPG) to reliably solve (hierarchical) structured sparsity problems and ensure sub-network validity; and (iii) automated sub-network construction using solutions from DHSPG/H2SPG and dependency graphs. Our empirical results demonstrate the efficacy of OTOv3 across various benchmarks in structured pruning and neural architecture search. OTOv3 produces sub-networks that match or exceed the state-of-the-arts. The source code will be available at //github.com/tianyic/only_train_once.
Convolutional neural networks have made significant progresses in edge detection by progressively exploring the context and semantic features. However, local details are gradually suppressed with the enlarging of receptive fields. Recently, vision transformer has shown excellent capability in capturing long-range dependencies. Inspired by this, we propose a novel transformer-based edge detector, \emph{Edge Detection TransformER (EDTER)}, to extract clear and crisp object boundaries and meaningful edges by exploiting the full image context information and detailed local cues simultaneously. EDTER works in two stages. In Stage I, a global transformer encoder is used to capture long-range global context on coarse-grained image patches. Then in Stage II, a local transformer encoder works on fine-grained patches to excavate the short-range local cues. Each transformer encoder is followed by an elaborately designed Bi-directional Multi-Level Aggregation decoder to achieve high-resolution features. Finally, the global context and local cues are combined by a Feature Fusion Module and fed into a decision head for edge prediction. Extensive experiments on BSDS500, NYUDv2, and Multicue demonstrate the superiority of EDTER in comparison with state-of-the-arts.
Approaches based on deep neural networks have achieved striking performance when testing data and training data share similar distribution, but can significantly fail otherwise. Therefore, eliminating the impact of distribution shifts between training and testing data is crucial for building performance-promising deep models. Conventional methods assume either the known heterogeneity of training data (e.g. domain labels) or the approximately equal capacities of different domains. In this paper, we consider a more challenging case where neither of the above assumptions holds. We propose to address this problem by removing the dependencies between features via learning weights for training samples, which helps deep models get rid of spurious correlations and, in turn, concentrate more on the true connection between discriminative features and labels. Extensive experiments clearly demonstrate the effectiveness of our method on multiple distribution generalization benchmarks compared with state-of-the-art counterparts. Through extensive experiments on distribution generalization benchmarks including PACS, VLCS, MNIST-M, and NICO, we show the effectiveness of our method compared with state-of-the-art counterparts.
Generative commonsense reasoning which aims to empower machines to generate sentences with the capacity of reasoning over a set of concepts is a critical bottleneck for text generation. Even the state-of-the-art pre-trained language generation models struggle at this task and often produce implausible and anomalous sentences. One reason is that they rarely consider incorporating the knowledge graph which can provide rich relational information among the commonsense concepts. To promote the ability of commonsense reasoning for text generation, we propose a novel knowledge graph augmented pre-trained language generation model KG-BART, which encompasses the complex relations of concepts through the knowledge graph and produces more logical and natural sentences as output. Moreover, KG-BART can leverage the graph attention to aggregate the rich concept semantics that enhances the model generalization on unseen concept sets. Experiments on benchmark CommonGen dataset verify the effectiveness of our proposed approach by comparing with several strong pre-trained language generation models, particularly KG-BART outperforms BART by 5.80, 4.60, in terms of BLEU-3, 4. Moreover, we also show that the generated context by our model can work as background scenarios to benefit downstream commonsense QA tasks.
Recurrent neural nets (RNN) and convolutional neural nets (CNN) are widely used on NLP tasks to capture the long-term and local dependencies, respectively. Attention mechanisms have recently attracted enormous interest due to their highly parallelizable computation, significantly less training time, and flexibility in modeling dependencies. We propose a novel attention mechanism in which the attention between elements from input sequence(s) is directional and multi-dimensional (i.e., feature-wise). A light-weight neural net, "Directional Self-Attention Network (DiSAN)", is then proposed to learn sentence embedding, based solely on the proposed attention without any RNN/CNN structure. DiSAN is only composed of a directional self-attention with temporal order encoded, followed by a multi-dimensional attention that compresses the sequence into a vector representation. Despite its simple form, DiSAN outperforms complicated RNN models on both prediction quality and time efficiency. It achieves the best test accuracy among all sentence encoding methods and improves the most recent best result by 1.02% on the Stanford Natural Language Inference (SNLI) dataset, and shows state-of-the-art test accuracy on the Stanford Sentiment Treebank (SST), Multi-Genre natural language inference (MultiNLI), Sentences Involving Compositional Knowledge (SICK), Customer Review, MPQA, TREC question-type classification and Subjectivity (SUBJ) datasets.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191