Catastrophic forgetting is a longstanding challenge in continual learning, where models lose knowledge from earlier tasks when learning new ones. While various mitigation strategies have been proposed for Multi-Layer Perceptrons (MLPs), recent architectural advances like Kolmogorov-Arnold Networks (KANs) have been suggested to offer intrinsic resistance to forgetting by leveraging localized spline-based activations. However, the practical behavior of KANs under continual learning remains unclear, and their limitations are not well understood. To address this, we present a comprehensive study of catastrophic forgetting in KANs and develop a theoretical framework that links forgetting to activation support overlap and intrinsic data dimension. We validate these analyses through systematic experiments on synthetic and vision tasks, measuring forgetting dynamics under varying model configurations and data complexity. Further, we introduce KAN-LoRA, a novel adapter design for parameter-efficient continual fine-tuning of language models, and evaluate its effectiveness in knowledge editing tasks. Our findings reveal that while KANs exhibit promising retention in low-dimensional algorithmic settings, they remain vulnerable to forgetting in high-dimensional domains such as image classification and language modeling. These results advance the understanding of KANs’ strengths and limitations, offering practical insights for continual learning system design.

As the gap between processor and memory performance widens, optimizing data access performance becomes increasingly critical. Hardware prefetching is a widely used technique to hide long-latency off-chip memory accesses, but state-of-the-art prefetchers struggle with diverse and dynamic access patterns. Their limited adaptability leads to excessive storage overhead and reduced effectiveness under memory-intensive workloads. We propose Zion, a comprehensive, adaptive, and lightweight hardware prefetcher for memory-intensive workloads. At its core, Zion uses Independent Temporal-Spatial Modules (ITSM) for broad pattern coverage and runtime adaptability to diverse memory access patterns. Moreover, Zion leverages runtime feedback to dynamically guide prefetching decisions and maintain efficiency under memory pressure. Extensive multi-core evaluations show that Zion consistently outperforms state-of-the-art prefetchers, achieving up to 43.2% performance improvement on SPEC and 43.0% on self-attention workloads, while maintaining low overhead and broad effectiveness.

Huixue Zhou, Hengrui Gu, Zaifu Zhan, Xi Liu, Kaixiong Zhou, Yongkang Xiao, Mingfu Liang, Srinivas Prasad Govindan, Piyush Chawla, Jiyan Yang, Xiangfei Meng, Huayu Li, Buyun Zhang, Liang Luo, Wen-Yen Chen, Yiping Han, Bo Long, Rui Zhang, Tianlong Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Densely structured pruning methods — which generate pruned models in a fully dense format, allowing immediate compression benefits without additional demands — are evolving due to their practical significance. Traditional techniques in this domain mainly revolve around coarser granularities, such as filter pruning, thereby limiting performance due to restricted pruning freedom. Recent advancements in Grouped Kernel Pruning (GKP) have enabled the utilization of finer granularities while maintaining a densely structured format. We observe that existing GKP methods often introduce dynamic operations to different aspects of their procedures at the cost of adding complications and/or imposing limitations (e.g., requiring an expensive mixture of clustering schemes), or contain dynamic pruning rates and sizes among groups that result in a reliance on custom architecture support for its pruned models. In this work, we argue that the best practice to introduce these dynamic operations to GKP is to make Conv2d(groups) (a.k.a. group count) flexible under an integral optimization, leveraging its ideal alignment with the infrastructure support of Grouped Convolution. Pursuing such a direction, we present a one-shot, post-train, data-agnostic GKP method that is more performant, adaptive, and efficient than its predecessors while simultaneously being user-friendly, with little-to-no hyper-parameter tuning or handcrafting of criteria required.

Large Vision-Language Models (VLMs) exhibit significant potential in multimodal tasks but often struggle with hallucinations—responses that are plausible yet visually ungrounded. In this work, we investigate the layer-wise prediction tendencies of VLMs and conduct an in-depth analysis of their decoding mechanism. We observe that VLMs tend to ``overthink'' during the final stages of decoding, making significant prediction shifts in the last few layers often favoring incorrect results, which leads to a surge in hallucinative outputs. Leveraging this localized pattern, we propose a novel decoding strategy inspired by the momentum analogy used in gradient descent-based optimizers. Our method enforces decoding consistency across layers in an adaptive manner during forward passes—an under-explored approach in existing works. This strategy significantly improves the reliability and performance of VLMs in various multimodal tasks, while introducing only negligible efficiency overhead.

Large language models (LLMs) can give out erroneous answers to factually rooted questions either as a result of undesired training outcomes or simply because the world has moved on after a certain knowledge cutoff date. Under such scenarios, *knowledge editing* often comes to the rescue by delivering efficient patches for such erroneous answers without significantly altering the rest, where many editing methods have seen reasonable success when the editing targets are simple and direct (e.g., *``what club does Lionel Messi currently play for?''*). However, knowledge fragments like this are often deeply intertwined in the real world, making effectively propagating the editing effect to non-directly related questions a practical challenge (to entertain an extreme example: [*"What car did the wife of the owner of the club that Messi currently plays for used to get to school in the 80s?"*](youtube.com/watch?v=DbwiHC1Fu-E\&t=132s)). Prior arts have coined this task as *multi-hop knowledge editing* with the most popular dataset being MQuAKE, serving as the sole evaluation benchmark for many later proposed editing methods due to the expensive nature of constructing knowledge editing datasets at scale. In this work, we reveal that **up to 33\% or 76\% of \mquake{}'s questions and ground truth labels are, in fact, corrupted in various fashions due to some unintentional clerical or procedural oversights**. Our work provides a detailed audit of MQuAKE's error pattern and a comprehensive fix without sacrificing its dataset capacity. Additionally, we benchmarked almost all proposed MQuAKE-evaluated editing methods on our post-fix dataset, **MQuAKE-Remastered**. We observe that many methods try to overfit the original MQuAKE by exploiting some dataset idiosyncrasies of MQuAKE. We provide a guideline on how to approach such datasets faithfully and show that a simple, minimally invasive approach — **GWalk** — can offer beyond SOTA editing performance without such exploitation. The MQuAKE-Remastered datasets and utilities are available at [huggingface.co/datasets/henryzhongsc/MQuAKE-Remastered](https://huggingface.co/datasets/henryzhongsc/MQuAKE-Remastered) and [github.com/henryzhongsc/MQuAKE-Remastered](https://github.com/henryzhongsc/MQuAKE-Remastered), respectively.

Generating novel enzymes for target molecules in zero-shot scenarios is a fundamental challenge in biomaterial synthesis and chemical production. Without known enzymes for a target molecule, training generative models becomes difficult due to the lack of direct supervision. To address this, we propose a retrieval-augmented generation method that uses existing enzyme-substrate data to guide enzyme design. Our method retrieves enzymes with substrates that share structural similarities with the target molecule, leveraging functional similarities in catalytic activity. Since none of the retrieved enzymes directly catalyze the target molecule, we use a conditioned discrete diffusion model to generate new enzymes based on the retrieved examples. An enzyme-substrate relationship classifier guides the generation process to ensure optimal protein sequence distributions. We evaluate our model on enzyme design tasks with diverse real-world substrates and show that it outperforms existing protein generation methods in catalytic capability, foldability, and docking accuracy. Additionally, we define the zero-shot substrate-specified enzyme generation task and introduce a dataset with evaluation benchmarks.

Deep learning models have attracted lots of research attention in time series classification (TSC) task in the past two decades. Recently, deep neural networks (DNN) have surpassed classical distance-based methods and achieved state-of-the-art performance. Despite their promising performance, deep neural networks (DNNs) have been shown to rely on spurious correlations present in the training data, which can hinder generalization. For instance, a model might incorrectly associate the presence of grass with the label "cat" if the training set have majority of cats lying in grassy backgrounds. However, the shortcut behavior of DNNs in time series remain under-explored. Most existing shortcut work are relying on external attributes such as gender, patients group, instead of focus on the internal bias behavior in time series models.In this paper, we take the first step to investigate and establish the point-based shortcut learning behavior in deep learning in time series classification task exists. We further propose a simple detection method based on other class to detect shortcut occurs without relying on test data or clean training classes. We test our proposed method in UCR time series datasets.

Yang Ouyang, Hengrui Gu, Shuhang Lin, Wenyue Hua, Jie Peng, Bhavya Kailkhura, Meijun Gao, Tianlong Chen, Kaixiong Zhou. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

The privacy concerns associated with the use of Large Language Models (LLMs) have grown dramatically with the development of pioneer LLMs such as ChatGPT. Differential Privacy (DP) techniques that utilize DP-SGD are explored in existing work to mitigate their privacy risks at the cost of generalization degradation. Our paper reveals that the flatness of DP-SGD trained models’ loss landscape plays an essential role in the trade-off between their privacy and generalization. We further propose a holistic framework Privacy-Flat to enforce appropriate weight flatness, which substantially improves model generalization with promising privacy protection. It innovates from three coarse-to-grained levels: Perturbation-aware min-max optimization within a layer, flatness-guided sparse prefix-tuning across layers, and weight knowledge distillation between private & non-private weights copies. We empirically demonstrate that our framework Privacy-Flat outperforms vanilla private training baseline while protecting privacy from membership inference attacks (MIA). Comprehensive experiments of both black-box and white-box scenarios are conducted to demonstrate the effectiveness of our proposal in enhancing generalization. The code link is provided at https://github.com/tiejin98/Privacy_ Flatness.

Multi-hop question answering (MQA) is one of the challenging tasks to evaluate machine's comprehension and reasoning abilities, where large language models (LLMs) have widely achieved the human-comparable performance.Due to the dynamics of knowledge facts in real world, knowledge editing has been explored to update model with the up-to-date facts while avoiding expensive re-training or fine-tuning.Starting from the edited fact, the updated model needs to provide cascading changes in the chain of MQA.The previous art simply adopts a mix-up prompt to instruct LLMs conducting multiple reasoning tasks sequentially, including question decomposition, answer generation, and conflict checking via comparing with edited facts.However, the coupling of these functionally-diverse reasoning tasks inhibits LLMs' advantages in comprehending and answering questions while disturbing them with the unskilled task of conflict checking.We thus propose a framework, Programmable knowledge editing for Multihop Question Answering (PokeMQA), to decouple the jobs.Specifically, we prompt LLMs to decompose knowledge-augmented multihop question, while interacting with a detached trainable scope detector to modulate LLMs behavior depending on external conflict signal.The experiments on three LLM backbones and two benchmark datasets validate our superiority in knowledge editing of MQA, outperforming all competitors by a large margin in almost all settings and consistently producing reliable reasoning process.Our code is available at https://github.com/Hengrui-Gu/PokeMQA.

Multivariate time series classification is an important task with widespread domains of applications. Recently, deep neural networks (DNN) have achieved state-of-the-art performance in time series classification. However, they often require large expert-labeled training datasets which can be infeasible in practice. In few-shot settings, i.e. only a limited number of samples per class are available in training data, DNNs show a significant drop in testing accuracy and poor generalization ability. In this paper, we propose to address these problems from an optimization and a loss function perspective. Specifically, we propose a new learning framework named COSCO consisting of a sharpness-aware minimization (SAM) optimization and a Prototypical loss function to improve the generalization ability of DNN for multivariate time series classification problems under few-shot setting. Our experiments demonstrate our proposed method outperforms the existing baseline methods. Our source code is available at: https://github.com/JRB9/COSCO.

Large Language Models (LLMs) have shown proficiency in question-answering tasks but often struggle to integrate real-time knowledge, leading to potentially outdated or inaccurate responses. This problem becomes even more challenging when dealing with multi-hop questions, since they require LLMs to update and integrate multiple knowledge pieces relevant to the questions. To tackle the problem, we propose the Retrieval-Augmented model Editing (RAE) framework for multi-hop question answering. RAE first retrieves edited facts and then refines the language model through in-context learning. Specifically, our retrieval approach, based on mutual information maximization, leverages the reasoning abilities of LLMs to identify chain facts that traditional similarity-based searches might miss. In addition, our framework includes a pruning strategy to eliminate redundant information from the retrieved facts, which enhances the editing accuracy and mitigates the hallucination problem. Our framework is supported by theoretical justification for its fact retrieval efficacy. Finally, comprehensive evaluation across various LLMs validates RAE's ability in providing accurate answers with updated knowledge. Our code is available at: https://github.com/sycny/RAE.

The premise for the great advancement of molecular machine learning is dependent on a considerable amount of labeled data. In many real-world scenarios, the labeled molecules are limited in quantity or laborious to derive. Recent pseudo-labeling methods are usually designed based on a single domain knowledge, thereby failing to understand the comprehensive molecular configurations and limiting their adaptability to generalize across diverse biochemical context. To this end, we introduce an innovative paradigm for dealing with the molecule pseudo-labeling, named as Molecular Data Programming (MDP). In particular, we adopt systematic supervision sources via crafting multiple graph labeling functions, which covers various molecular structural knowledge of graph kernels, molecular fingerprints, and topological features. Each of them creates an uncertain and biased labels for the unlabeled molecules. To address the decision conflicts among the diverse pseudo-labels, we design a label synchronizer to differentiably model confidences and correlations between the labeling functions, which yields probabilistic molecular labels to adapt for specific applications. These probabilistic molecular labels are used to train a molecular classifier for improving its generalization capability. On eight bench-mark datasets, we empirically demonstrate the effectiveness of MDP on the weakly supervised molecule classification tasks, achieving an average improvement of 9.5%. The code is in: https://github.com/xinjuan1/MDP/.

Sharpness-aware minimization (SAM) has received increasing attention in computer vision since it can effectively eliminate the sharp local minima from the training trajectory and mitigate generalization degradation. However, SAM requires two sequential gradient computations during the optimization of each step: one to obtain the perturbation gradient and the other to obtain the updating gradient. Compared with the base optimizer (e.g., Adam), SAM doubles the time overhead due to the additional perturbation gradient. By dissecting the theory of SAM and observing the training gradient of the molecular graph transformer, we propose a new algorithm named GraphSAM, which reduces the training cost of SAM and improves the generalization performance of graph transformer models. There are two key factors that contribute to this result: (i) \textit{gradient approximation}: we use the updating gradient of the previous step to approximate the perturbation gradient at the intermediate steps smoothly (\textbf{increases efficiency}); (ii) \textit{loss landscape approximation}: we theoretically prove that the loss landscape of GraphSAM is limited to a small range centered on the expected loss of SAM (\textbf{guarantees generalization performance}). The extensive experiments on six datasets with different tasks demonstrate the superiority of GraphSAM, especially in optimizing the model update process.

Knowledge Graphs (KGs) are potent frameworks for knowledge representation and reasoning. Nevertheless, KGs are inherently incomplete, leaving numerous uncharted relationships and facts awaiting discovery. Deep learning methodologies have proven effective in enhancing KG completion by framing it as a link prediction task, where the goal is to discern the validity of a triple comprising a head, relation, and tail. The significance of structural information in assessing the validity of a triple within a KG is well-established. However, quantifying this structural information poses a challenge. We need to pinpoint the metric that encapsulates the structural information of a triple and smoothly incorporate this metric into the link prediction learning process. In this study, we recognize the critical importance of the intersection among the $k$-hop neighborhoods of the head, relation, and tail when determining the validity of a triple. To address this, we introduce a novel randomized algorithm designed to efficiently generate intersection features for candidate triples. Our experimental results demonstrate that a straightforward fully-connected network leveraging these intersection features can surpass the performance of established KG embedding models and even outperform graph neural network baselines. Additionally, we highlight the substantial training time efficiency gains achieved by our network trained on intersection features.

Graph neural networks (GNNs) have exhibited prominent performance in learning graph-structured data. Considering node classification task, based on the i.i.d assumption among node labels, the traditional supervised learning simply sums up cross-entropy losses of the independent training nodes and applies the average loss to optimize GNNs’ weights. But different from other data formats, the nodes are naturally connected. It is found that the independent distribution modeling of node labels restricts GNNs’ capability to generalize over the entire graph and defend adversarial attacks. In this work, we propose a new framework, termed joint-cluster supervised learning, to model the joint distribution of each node with its corresponding cluster. We learn the joint distribution of node and cluster labels conditioned on their representations, and train GNNs with the obtained joint loss. In this way, the data-label reference signals extracted from the local cluster explicitly strengthen the discrimination ability on the target node. The extensive experiments demonstrate that our joint-cluster supervised learning can effectively bolster GNNs’ node classification accuracy. Furthermore, being benefited from the reference signals which may be free from spiteful interference, our learning paradigm significantly protects the node classification from being affected by the adversarial attack.

Explainable machine learning significantly improves the transparency of deep neural networks. However, existing work is constrained to explaining the behavior of individual model predictions, and lacks the ability to transfer the explanation across various models and tasks. This limitation results in explaining various tasks being time- and resource-consuming. To address this problem, we introduce a Transferable Vision Explainer (TVE) that can effectively explain various vision models in downstream tasks. Specifically, the transferability of TVE is realized through a pre-training process on large-scale datasets towards learning the meta-attribution. This meta-attribution leverages the versatility of generic backbone encoders to comprehensively encode the attribution knowledge for the input instance, which enables TVE to seamlessly transfer to explaining various downstream tasks, without the need for training on task-specific data. Empirical studies involve explaining three different architectures of vision models across three diverse downstream datasets. The experiment results indicate TVE is effective in explaining these tasks without the need for additional training on downstream data.

Model compression is one of the most popular approaches to improve the accessibility of Large Language Models (LLMs) by reducing their memory footprint. However, the gaining of such efficiency benefits often simultaneously demands extensive engineering efforts and intricate designs to mitigate the performance decline. In this work, we leverage *(Soft) Prompt Tuning* in its most vanilla form and discover such conventionally learned soft prompts can recover the performance of compressed LLMs. More surprisingly, we observe such recovery effect to be transferable among different tasks and models (albeit natural tokenizer and dimensionality limitations), resulting in further overhead reduction and yet, subverting the common belief that learned soft prompts are task-specific. Our work is fully orthogonal and compatible with model compression frameworks such as pruning and quantization, where we enable up to $8\times$ compressed LLM (with a joint 4-bit quantization and 50% weight pruning compression) to match its uncompressed counterparts on popular benchmarks. We note that we are the first to reveal vanilla Parameter-Efficient Fine-Tuning (PEFT) techniques have the potential to be utilized under a compression recovery context, opening a new line of opportunities for model accessibility advancement while freeing our fellow researchers from the previously present engineering burdens and constraints. The code is available at https://github.com/zirui-ray-liu/compress-then-prompt.

Suppose a self-driving car is crashing into pedestrians, or a chatbot is instructing its users to conduct criminal wrongdoing; the stakeholders of such products will undoubtedly want to patch these catastrophic errors as soon as possible. To address such concerns, Model Editing: the study of efficiently patching model behaviors without significantly altering their general performance, has seen considerable activity, with hundreds of editing techniques developed in various domains such as CV and NLP. However, the graph learning community has objectively fallen behind with only a few Graph Neural Network-compatible — and just one GNN-specific — model editing methods available, where all of which are limited in their practical scope. We argue that the impracticality of these methods lies in their lack of Sequential Editing Robustness: the ability to edit multiple errors sequentially, and therefore fall short in effectiveness, as this approach mirrors how errors are discovered and addressed in the real world. In this paper, we delve into the specific reasons behind the difficulty of editing GNNs in succession and observe the root cause to be model overfitting. We subsequently propose a simple yet effective solution — SEED-GNN — by leveraging overfit-prevention techniques in a GNN-specific context to derive the first and only GNN model editing method that scales practically. Additionally, we formally frame the task paradigm of GNN editing and hope to inspire future research in this crucial but currently overlooked field. Please refer to our GitHub repository for code and checkpoints.

Despite the recent progress of molecular representation learning, its effectiveness is assumed on the close-world assumptions that training and testing graphs are from identical distribution. The open-world test dataset is often mixed with out-of-distribution (OOD) samples, where the deployed models will struggle to make accurate predictions. The misleading estimations of molecules' properties in drug screening or design can result in the tremendous waste of wet-lab resources and delay the discovery of novel therapies. Traditional detection methods need to trade off OOD detection and in-distribution (ID) classification performance since they share the same representation learning model. In this work, we propose to detect OOD molecules by adopting an auxiliary diffusion model-based framework, which compares similarities between input molecules and reconstructed graphs. Due to the generative bias towards reconstructing ID training samples, the similarity scores of OOD molecules will be much lower to facilitate detection. Although it is conceptually simple, extending this vanilla framework to practical detection applications is still limited by two significant challenges. First, the popular similarity metrics based on Euclidian distance fail to consider the complex graph structure. Second, the generative model involving iterative denoising steps is notoriously time-consuming especially when it runs on the enormous pool of drugs. To address these challenges, our research pioneers an approach of Prototypical Graph Reconstruction for Molecular OOd Detection, dubbed as PGR-MOOD. Specifically, PGR-MOOD hinges on three innovations: i) An effective metric to comprehensively quantify the matching degree of input and reconstructed molecules according to their discrete edges and continuous node features; ii) A creative graph generator to construct a list of prototypical graphs that are in line with ID distribution but away from OOD one; iii) An efficient and scalable OOD detector to compare the similarity between test samples and pre-constructed prototypical graphs and omit the generative process on every new molecule. Extensive experiments on ten benchmark datasets and six baselines are conducted to demonstrate our superiority: PGR-MOOD achieves more than 8% of average improvement in terms of detection AUC and AUPR accompanied by the reduced cost of testing time and memory consumption. The anonymous code is in: https://github.com/se7esx/PGR-MOOD.

Deep neural networks are ubiquitously adopted in many applications, such as computer vision, natural language processing, and graph analytics. However, well-trained neural networks can make prediction errors after deployment as the world changes. \textit{Model editing} involves updating the base model to correct prediction errors with less accessible training data and computational resources. Despite recent advances in model editors in computer vision and natural language processing, editable training in graph neural networks (GNNs) is rarely explored. The challenge with editable GNN training lies in the inherent information aggregation across neighbors, which can lead model editors to affect the predictions of other nodes unintentionally. In this paper, we first observe the gradient of cross-entropy loss for the target node and training nodes with significant inconsistency, which indicates that directly fine-tuning the base model using the loss on the target node deteriorates the performance on training nodes. Motivated by the gradient inconsistency observation, we propose a simple yet effective \underline{G}radient \underline{R}ewiring method for \underline{E}ditable graph neural network training, named \textbf{GRE}. Specifically, we first store the anchor gradient of the loss on training nodes to preserve the locality. Subsequently, we rewire the gradient of the loss on the target node to preserve performance on the training node using anchor gradient. Experiments demonstrate the effectiveness of GRE on various model architectures and graph datasets in terms of multiple editing situations. The source code is available at \url{https://github.com/zhimengj0326/Gradient_rewiring_editing}.

The exponential growth in scholarly publications necessitates advanced tools for efficient article retrieval, especially in interdisciplinary fields where diverse terminologies are used to describe similar research. Traditional keyword-based search engines often fall short in assisting users who may not be familiar with specific terminologies. To address this, we present a knowledge graph based paper search engine for biomedical research to enhance the user experience in discovering relevant queries and articles. The system, dubbed DiscoverPath, employs Named Entity Recognition (NER) and part-of-speech (POS) tagging to extract terminologies and relationships from article abstracts to create a KG. To reduce information overload, DiscoverPath presents users with a focused subgraph containing the queried entity and its neighboring nodes and incorporates a query recommendation system enabling users to iteratively refine their queries. The system is equipped with an accessible Graphical User Interface that provides an intuitive visualization of the KG, query recommendations, and detailed article information, enabling efficient article retrieval, thus fostering interdisciplinary knowledge exploration. DiscoverPath is open-sourced at https://github.com/ynchuang/DiscoverPath with a demo video at Youtube.

Training graph neural networks (GNNs) is extremely time consuming because sparse graph-based operations are hard to be accelerated by community hardware. Prior art successfully reduces the computation cost of dense matrix based operations (e.g., convolution and linear) via sampling-based approximation. However, unlike dense matrices, sparse matrices are stored in the irregular data format such that each row/column may have different number of non-zero entries. Thus, compared to the dense counterpart, approximating sparse operations has two unique challenges (1) we cannot directly control the efficiency of approximated sparse operation since the computation is only executed on non-zero entries; (2) sampling sparse matrices is much more inefficient due to the irregular data format. To address the issues, our key idea is to control the accuracy-efficiency trade off by optimizing computation resource allocation layer-wisely and epoch-wisely. For the first challenge, we customize the computation resource to different sparse operations, while limit the total used resource below a certain budget. For the second challenge, we cache previous sampled sparse matrices to reduce the epoch-wise sampling overhead. Finally, we propose a switching mechanisms to improve the generalization of GNNs trained with approximated operations. To this end, we propose Randomized Sparse Computation. In practice, rsc can achieve up to 11.6X speedup for a single sparse operation and 1.6X end-to-end wall-clock time speedup with almost no accuracy drop.

Transformer-based models are powerful for modeling temporal dynamics of user preference in sequential recommendation. Most of the variants adopt the Softmax transformation in the self-attention layers to generate dense attention probabilities. However, real-world item sequences are often noisy, containing a mixture of true-positive and false-positive interactions. Such dense attentions inevitably assign probability mass to noisy or irrelevant items, leading to sub-optimal performance and poor explainability. Here we propose a Probabilistic Masked Attention Network (PMAN) to identify the sparse pattern of attentions, which is more desirable for pruning noisy items in sequential recommendation. Specifically, we employ a probabilistic mask to achieve sparse attentions under a constrained optimization framework. As such, PMAN allows to select which information is critical to be retained or dropped in a data-driven fashion. Experimental studies on real-world benchmark datasets show that PMAN is able to improve the performance of Transformers significantly.

As the model size grows rapidly, fine-tuning the large pre-trained language model has become increasingly difficult due to its extensive memory usage. Previous works usually focus on reducing the number of trainable parameters in the network. While the model parameters do contribute to memory usage, the primary memory bottleneck during training arises from storing feature maps, also known as activations, as they are crucial for gradient calculation. Notably, machine learning models are typically trained using stochastic gradient descent. We argue that in stochastic optimization, models can handle noisy gradients as long as the gradient estimator is unbiased with reasonable variance. Following this motivation, we propose a new family of unbiased estimators called \sas, for matrix production with reduced variance, which only requires storing the sub-sampled activations for calculating the gradient. Our work provides both theoretical and experimental evidence that, in the context of tuning transformers, our proposed estimators exhibit lower variance compared to existing ones. By replacing the linear operation with our approximated one in transformers, we can achieve up to 2.7X peak memory reduction with almost no accuracy drop and enables up to $6.4\times$ larger batch size. Under the same hardware, \sas enables better down-streaming task performance by applying larger models and/or faster training speed with larger batch sizes. The code is available at https://anonymous.4open.science/r/WTACRS-A5C5/.

Graph Neural Networks (GNNs) have achieved state-of-the-art performance in recommender systems. Nevertheless, the process of searching and ranking from a large item corpus usually requires high latency, which limits the widespread deployment of GNNs in industry-scale applications. To address this issue, many methods compress user/item representations into the binary embedding space to reduce space requirements and accelerate inference. Also, they use the Straight-through Estimator (STE) to prevent vanishing gradients during back-propagation. However, the STE often causes the gradient mismatch problem, leading to sub-optimal results.

Time series anomaly detection is a challenging task with a wide range of real-world applications. Due to label sparsity, training a deep anomaly detector often relies on unsupervised approaches. Recent efforts have been devoted to time series domain adaptation to leverage knowledge from similar domains. However, existing solutions may suffer from negative knowledge transfer on anomalies due to their diversity and sparsity. Motivated by the empirical study of context alignment between two domains, we aim to transfer knowledge between two domains via adaptively sampling context information for two domains. This is challenging because it requires simultaneously modeling the complex in-domain temporal dependencies and cross-domain correlations while exploiting label information from the source domain. To this end, we propose a framework that combines context sampling and anomaly detection into a joint learning procedure. We formulate context sampling into the Markov decision process and exploit deep reinforcement learning to optimize the time series domain adaptation process via context sampling and design a tailored reward function to generate domain-invariant features that better align two domains for anomaly detection. Experiments on three public datasets show promise for knowledge transfer between two similar domains and two entirely different domains.

Graph neural networks (GNNs), which learn the node representations by recursively aggregating information from its neighbors, have become a predominant computational tool in many domains. To handle large-scale graphs, most of the existing methods partition the input graph into multiple sub-graphs (e.g., through node clustering) and apply batch training to save memory cost. However, such batch training will lead to label bias within each batch and result in over- confidence in model predictions. Since the connected nodes with positively related labels tend to be assigned together, the traditional cross-entropy minimization process attends on the predictions of biased classes at a batch to intensify the overfitting issue. To overcome the problem of label bias, we propose adaptive label smoothing (ALS) method to replace the one-hot hard labels with smoothed ones, which learns to allocate label confidences from the biased classes to the others. Specifically, ALS propagates node labels to aggregate the neighborhood label distribution in a pre-processing step, and then updates the optimal smoothed labels online to adapt to specific graph structure. Experiments on the real-world datasets demonstrate that ALS can be generally applied to the main scalable learning frameworks to calibrate the biased labels and improve generalization performances.

Training deep graph neural networks (GNNs) is notoriously hard. Besides the standard plights in training deep architectures such as vanishing gradients and overfitting, it also uniquely suffers from over-smoothing, information squashing, and so on, which limits their potential power for encoding the high-order neighbor structure in large-scale graphs. Although numerous efforts are proposed to address these limitations, such as various forms of skip connections, graph normalization, and random dropping, it is difficult to disentangle the advantages brought by a deep GNN architecture from those "tricks" necessary to train such an architecture. Moreover, the lack of a standardized benchmark with fair and consistent experimental settings poses an almost insurmountable obstacle to gauge the effectiveness of new mechanisms. In view of those, we present the first fair and reproducible benchmark dedicated to assessing the "tricks" of training deep GNNs. We categorize existing approaches, investigate their hyperparameter sensitivity, and unify the basic configuration. Comprehensive evaluations are then conducted on tens of representative graph datasets including the recent large-scale Open Graph Benchmark, with diverse deep GNN backbones. We demonstrate that an organic combo of initial connection, identity mapping, group and batch normalization attains the new state-of-the-art results for deep GNNs on large datasets. Codes are available: https://github.com/VITA-Group/Deep_GCN_Benchmarking.

Graph neural networks (GNNs) have received tremendous attention due to their superiority in learning node representations. These models rely on message passing and feature transformation functions to encode the structural and feature information from neighbors. However, stacking more convolutional layers significantly decreases the performance of GNNs. Most recent studies attribute this limitation to the over-smoothing issue, where node embeddings converge to indistinguishable vectors. Through a number of experimental observations, we argue that the main factor degrading the performance is the unstable forward normalization and backward gradient resulted from the improper design of the feature transformation, especially for shallow GNNs where the over-smoothing has not happened. Therefore, we propose a novel orthogonal feature transformation, named Ortho-GConv, which could generally augment the existing GNN backbones to stabilize the model training and improve the model's generalization performance. Specifically, we maintain the orthogonality of the feature transformation comprehensively from three perspectives, namely hybrid weight initialization, orthogonal transformation, and orthogonal regularization. By equipping the existing GNNs (e.g. GCN, JKNet, GCNII) with Ortho-GConv, we demonstrate the generality of the orthogonal feature transformation to enable stable training, and show its effectiveness for node and graph classification tasks.

Training graph neural networks (GNNs) with good generalizability on large-scale graphs is a challenging problem. Existing methods mainly divide the input graph into multiple subgraphs and train them in different batches to improve training scalability. However, the local batches obtained by such a strategy could contain topological bias compared with the complete graph structure. It has been studied that the topological bias results in more significant gaps between training and testing performances, or worse generalization robustness. A straightforward solution is to utilize contrastive learning, and train node embeddings to be robust and invariant among the augmented imperfect graphs. However, most of the existing work are inefficient by contrasting extensive node pairs at the large-scale graph. With random data augmentation, they may deteriorate the embedding process by transforming well-sampled batches into meaningless graph structures.

Learning useful interactions between input features is crucial for tabular data modeling. Recent efforts start to explicitly model the feature interactions with graph, where each feature is treated as an individual node. However, the existing graph construction methods either heuristically formulate a fixed feature-interaction graph based on specific domain knowledge, or simply apply attention function to compute the pairwise feature similarities for each sample. While the fixed graph may be sub-optimal to downstream tasks, the sample-wise graph construction is time-consuming during model training and inference. To tackle these issues, we propose a framework named Table2Graph to transform the feature interaction modeling to learning a unified graph. Represented as a probability adjacency matrix, the unified graph learns to model the key feature interactions shared by the diverse samples in the tabular data. To well optimize the unified graph, we employ the reinforcement learning policy to capture the key feature interactions stably. A sparsity constraint is also proposed to regularize the learned graph from being overly-sparse/smooth. The experimental results in a variety of real-world applications demonstrate the effectiveness and efficiency of our Table2Graph, in terms of the prediction accuracy and feature interaction detection.

Despite the promising representation learning of graph neural networks (GNNs), the supervised training of GNNs notoriously requires large amounts of labeled data from each application. An effective solution is to apply the transfer learning in graph: using easily accessible information to pre-train GNNs, and fine-tuning them to optimize the downstream task with only a few labels. Recently, many efforts have been paid to design the self-supervised pretext tasks, and encode the universal graph knowledge among the various applications. However, they rarely notice the inherent training objective gap between the pretext and downstream tasks. This significant gap often requires costly fine-tuning for adapting the pre-trained model to downstream problem, which prevents the efficient elicitation of pre-trained knowledge and then results in poor results. Even worse, the naive pre-training strategy usually deteriorates the downstream task, and damages the reliability of transfer learning in graph data. To bridge the task gap, we propose a novel transfer learning paradigm to generalize GNNs, namely graph pre-training and prompt tuning (GPPT). Specifically, we first adopt the masked edge prediction, the most simplest and popular pretext task, to pre-train GNNs. Based on the pre-trained model, we propose the graph prompting function to modify the standalone node into a token pair, and reformulate the downstream node classification looking the same as edge prediction. The token pair is consisted of candidate label class and node entity. Therefore, the pre-trained GNNs could be applied without tedious fine-tuning to evaluate the linking probability of token pair, and produce the node classification decision. The extensive experiments on eight benchmark datasets demonstrate the superiority of GPPT, delivering an average improvement of 4.29% in few-shot graph analysis and accelerating the model convergence up to 4.32X. The code is available in: https://github.com/MingChen-Sun/GPPT.

Large-scale graph training is a notoriously challenging problem for graph neural networks (GNNs). Due to the nature of evolving graph structures into the training process, vanilla GNNs usually fail to scale up, limited by the GPU memory space. Up to now, though numerous scalable GNN architectures have been proposed, we still lack a comprehensive survey and fair benchmark of this reservoir to find the rationale for designing scalable GNNs. To this end, we first systematically formulate the representative methods of large-scale graph training into several branches and further establish a fair and consistent benchmark for them by a greedy hyperparameter searching. In addition, regarding efficiency, we theoretically evaluate the time and space complexity of various branches and empirically compare them w.r.t GPU memory usage, throughput, and convergence. Furthermore, We analyze the pros and cons for various branches of scalable GNNs and then present a new ensembling training manner, named EnGCN, to address the existing issues. Remarkably, our proposed method has achieved new state-of-the-art (SOTA) performance on large-scale datasets. Our code is available at https://github.com/VITA-Group/LargeScaleGCN_Benchmarking.

There has been an explosion of interest in designing various Knowledge Graph Neural Networks (KGNNs), which achieve state-of-the-art performance and provide great explainability for recommendation. The promising performance is mainly resulting from their capability of capturing high-order proximity messages over the knowledge graphs. However, training KGNNs at scale is challenging due to the high memory usage. In the forward pass, the automatic differentiation engines (e.g., TensorFlow/PyTorch) generally need to cache all intermediate activation maps in order to compute gradients in the backward pass, which leads to a large GPU memory footprint. Existing work solves this problem by utilizing multi-GPU distributed frameworks. Nonetheless, this poses a practical challenge when seeking to deploy KGNNs in memory-constrained environments, especially for industry-scale graphs.

We study time-series classification (TSC), a fundamental task of time-series data mining. Prior work has approached TSC from two major directions: (1) similarity-based methods that classify time-series based on the nearest neighbors, and (2) deep learning models that directly learn the representations for classification in a data-driven manner. Motivated by the different working mechanisms within these two research lines, we aim to connect them in such a way as to jointly model time-series similarities and learn the representations. This is a challenging task because it is unclear how we should efficiently leverage similarity information. To tackle the challenge, we propose Similarity-Aware Time-Series Classification (SimTSC), a conceptually simple and general framework that models similarity information with graph neural networks (GNNs). Specifically, we formulate TSC as a node classification problem in graphs, where the nodes correspond to time-series, and the links correspond to pair-wise similarities. We further design a graph construction strategy and a batch training algorithm with negative sampling to improve training efficiency. We instantiate SimTSC with ResNet as the backbone and Dynamic Time Warping (DTW) as the similarity measure. Extensive experiments on the full UCR datasets and several multivariate datasets demonstrate the effectiveness of incorporating similarity information into deep learning models in both supervised and semi-supervised settings. Our code is available at https://github.com/daochenzha/SimTSC.

Graph Neural Networks (GNNs) provide a class of powerful architectures that are effective for graph-based collaborative filtering. Nevertheless, GNNs are known to be vulnerable to adversarial perturbations. Adversarial training is a simple yet effective way to improve the robustness of neural models. For example, many prior studies inject adversarial perturbations into either node features or hidden layers of GNNs. However, perturbing graph structures has been far less studied in recommendations.

Anomaly detection is an important data mining task with numerous applications, such as intrusion detection, credit card fraud detection, and video surveillance. However, given a specific complicated task with complicated data, the process of building an effective deep learning-based system for anomaly detection still highly relies on human expertise and laboring trials. Also, while neural architecture search (NAS) has shown its promise in discovering effective deep architectures in various domains, such as image classification, object detection, and semantic segmentation, contemporary NAS methods are not suitable for anomaly detection due to the lack of intrinsic search space, unstable search process, and low sample efficiency. To bridge the gap, in this article, we propose AutoAD, an automated anomaly detection framework, which aims to search for an optimal neural network model within a predefined search space. Specifically, we first design a curiosity-guided search strategy to overcome the curse of local optimality. A controller, which acts as a search agent, is encouraged to take actions to maximize the information gain about the controller's internal belief. We further introduce an experience replay mechanism based on self-imitation learning to improve the sample efficiency. Experimental results on various real-world benchmark datasets demonstrate that the deep model identified by AutoAD achieves the best performance, comparing with existing handcrafted models and traditional search methods.

Human-designed data augmentation strategies have been replaced by automatically learned augmentation policy in the past two years. Specifically, recent work has empirically shown that the superior performance of the automated data augmentation methods stems from increasing the diversity of augmented data [4], [5]. However, two factors regarding the diversity of augmented data are still missing: 1) the explicit definition (and thus measurement) of diversity and 2) the quantifiable relationship between diversity and its regularization effects. To bridge this gap, we propose a diversity measure called Variance Diversity and theoretically show that the regularization effect of data augmentation is promised by Variance Diversity. We validate in experiments that the relative gain from automated data augmentation in test accuracy is highly correlated to Variance Diversity. An unsupervised sampling-based framework, DivAug, is designed to directly maximize Variance Diversity and hence strengthen the regularization effect. Without requiring a separate search process, the performance gain from DivAug is comparable with the state-of-the-art method with better efficiency. Moreover, under the semi-supervised setting, our framework can further improve the performance of semi-supervised learning algorithms compared to RandAugment, making it highly applicable to real-world problems, where labeled data is scarce. The code is available at https://github.com/warai-0toko/DivAug.

Outlier detection is an important data mining task with numerous applications such as intrusion detection, credit card fraud detection, and video surveillance. However, given a specific task with complex data, the process of building an effective deep learning based system for outlier detection still highly relies on human expertise and laboring trials. Moreover, while Neural Architecture Search (NAS) has shown its promise in discovering effective deep architectures in various domains, such as image classification, object detection and semantic segmentation, contemporary NAS methods are not suitable for outlier detection due to the lack of intrinsic search space and low sample efficiency. To bridge the gap, in this paper, we propose AutoOD, an automated outlier detection framework, which aims to search for an optimal neural network model within a predefined search space. Specifically, we introduce an experience replay mechanism based on self-imitation learning to improve the sample efficiency. Experimental results on various real-world benchmark datasets demonstrate that the deep model identified by AutoOD achieves the best performance, comparing with existing handcrafted models and traditional search methods.

Graph neural networks (GNNs) integrate deep architectures and topological structure modeling in an effective way. However, the performance of existing GNNs would decrease significantly when they stack many layers, because of the over-smoothing issue. Node embeddings tend to converge to similar vectors when GNNs keep recursively aggregating the representations of neighbors. To enable deep GNNs, several methods have been explored recently. But they are developed from either techniques in convolutional neural networks or heuristic strategies. There is no generalizable and theoretical principle to guide the design of deep GNNs. To this end, we analyze the bottleneck of deep GNNs by leveraging the Dirichlet energy of node embeddings, and propose a generalizable principle to guide the training of deep GNNs. Based on it, a novel deep GNN framework -- Energetic Graph Neural Networks (EGNN) is designed. It could provide lower and upper constraints in terms of Dirichlet energy at each layer to avoid over-smoothing. Experimental results demonstrate that EGNN achieves state-of-the-art performance by using deep layers.

Session-based recommendation aims to predict the next item that is most likely to be clicked by an anonymous user, based on his/her clicking sequence within one visit. It becomes an essential function of many recommender systems since it protects privacy. However, as the accumulated session records keep increasing, it becomes challenging to model the user interests since they would drift when the time span is large. Efforts have been devoted to handling dynamic user interests by modeling all historical sessions at one time or conducting offline retraining regularly. These solutions are far from practical requirements in terms of efficiency and capturing timely user interests. To this end, we propose a memory-efficient framework - TASRec. It constructs a graph for each day to model the relations among items. Thus, the same item on different days could have different neighbors, corresponding to the drifting user interests. We design a tailored graph neural network to embed this dynamic graph of items and learn temporal augmented item representations. Based on this, we leverage a sequential neural architecture to predict the next item of a given sequence. Experiments on real-world datasets demonstrate that TASRec outperforms state-of-the-art session-based recommendation methods.

The classification of graph-structured data has be-come increasingly crucial in many disciplines. It has been observed that the implicit or explicit hierarchical community structures preserved in real-world graphs could be useful for downstream classification applications. A straightforward way to leverage the hierarchical structure is to make use the pooling algorithms to cluster nodes into fixed groups, and shrink the input graph layer by layer to learn the pooled graphs.However, the pool shrinking discards the graph details to make it hard to distinguish two non-isomorphic graphs, and the fixed clustering ignores the inherent multiple characteristics of nodes. To compensate the shrinking loss and learn the various nodes’ characteristics, we propose the multi-channel graph neural networks (MuchGNN). Motivated by the underlying mechanisms developed in convolutional neural networks, we define the tailored graph convolutions to learn a series of graph channels at each layer, and shrink the graphs hierarchically to en-code the pooled structures. Experimental results on real-world datasets demonstrate the superiority of MuchGNN over the state-of-the-art methods.

Graph data are pervasive in many real-world applications. Recently, increasing attention has been paid on graph neural networks (GNNs), which aim to model the local graph structures and capture the hierarchical patterns by aggregating the information from neighbors with stackable network modules. Motivated by the observation that different nodes often require different iterations of aggregation to fully capture the structural information, in this paper, we propose to explicitly sample diverse iterations of aggregation for different nodes to boost the performance of GNNs. It is a challenging task to develop an effective aggregation strategy for each node, given complex graphs and sparse features. Moreover, it is not straightforward to derive an efficient algorithm since we need to feed the sampled nodes into different number of network layers. To address the above challenges, we propose Policy-GNN, a meta-policy framework that models the sampling procedure and message passing of GNNs into a combined learning process. Specifically, Policy-GNN uses a meta-policy to adaptively determine the number of aggregations for each node. The meta-policy is trained with deep reinforcement learning~(RL) by exploiting the feedback from the model. We further introduce parameter sharing and a buffer mechanism to boost the training efficiency. Experimental results on three real-world benchmark datasets suggest that Policy-GNN significantly outperforms the state-of-the-art alternatives, showing the promise in aggregation optimization for GNNs.

Detecting statistical interactions between input features is a crucial and challenging task. Recent advances demonstrate that it is possible to extract learned interactions from trained neural networks. It has also been observed that, in neural networks, any interacting features must follow a strongly weighted connection to common hidden units. Motivated by the observation, in this paper, we propose to investigate the interaction detection problem from a novel topological perspective by analyzing the connectivity in neural networks. Specially, we propose a new measure for quantifying interaction strength, based upon the well-received theory of persistent homology. Based on this measure, a Persistence Interaction Dection (PID) algorithm is developed to efficiently detect interactions. Our proposed algorithm is evaluated across a number of interaction detection tasks on several synthetic and real-world datasets with different hyperparameters. Experimental results validate that the PID algorithm outperforms the state-of-the-art baselines.

Graph neural networks (GNNs), which learn the representation of a node by aggregating its neighbors, have become an effective computational tool in downstream applications. Over-smoothing is one of the key issues which limit the performance of GNNs as the number of layers increases. It is because the stacked aggregators would make node representations converge to indistinguishable vectors. Several attempts have been made to tackle the issue by bringing linked node pairs close and unlinked pairs distinct. However, they often ignore the intrinsic community structures and would result in sub-optimal performance. The representations of nodes within the same community/class need be similar to facilitate the classification, while different classes are expected to be separated in embedding space. To bridge the gap, we introduce two over-smoothing metrics and a novel technique, i.e., differentiable group normalization (DGN). It normalizes nodes within the same group independently to increase their smoothness, and separates node distributions among different groups to significantly alleviate the over-smoothing issue. Experiments on real-world datasets demonstrate that DGN makes GNN models more robust to over-smoothing and achieves better performance with deeper GNNs.

Deep neural network (DNN) has become an effective computational tool because of its superior performance in practice. However, the generalization of DNN still largely depends on the training data, no matter in quantity or quality. In this paper, we propose a knowledge instillation framework, named NeuKI, for feed-forward DNN, aiming to enhance learning performance with the aid of knowledge. This task is particularly challenging due to the complicated nature of knowledge and numerous variants of DNN architectures. To bridge the gap, we construct a separate knowledge-DNN faithfully encoding the instilled knowledge for joint training. The core idea is to regularize the training of target-DNN with the constructed knowledge-DNN, so that the instilled knowledge can guide the model training. The proposed NeuKI is demonstrated to be applicable to both knowledge rules and constraints, where rules are encoded by structure and constraints are handled by loss. Experiments are conducted on several real-world datasets from different domains, and the results demonstrate the effectiveness of NeuKI in improving learning performance, as well as relevant data efficiency and model interpretability.

Experience replay enables reinforcement learning agents to memorize and reuse past experiences, just as humans replay memories for the situation at hand. Contemporary off-policy algorithms either replay past experiences uniformly or utilize a rule-based replay strategy, which may be sub-optimal. In this work, we consider learning a replay policy to optimize the cumulative reward. Replay learning is challenging because the replay memory is noisy and large, and the cumulative reward is unstable. To address these issues, we propose a novel experience replay optimization (ERO) framework which alternately updates two policies: the agent policy, and the replay policy. The agent is updated to maximize the cumulative reward based on the replayed data, while the replay policy is updated to provide the agent with the most useful experiences. The conducted experiments on various continuous control tasks demonstrate the effectiveness of ERO, empirically showing promise in experience replay learning to improve the performance of off-policy reinforcement learning algorithms.