Jianzhu Bao, Haozhen Zhang, Kuicai Dong, Bozhi Wu, Sarthak Ketanbhai Modi, Zi Pong Lim, Yon Shin Teo, Wenya Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Black-box prompt tuning has become a prevalent parameter-efficient paradigm that leverages the capabilities of large language models (LLMs) for customized applications in specific downstream tasks. In practical scenarios, downstream tasks frequently involve data distributions that are heavily imbalanced. Such imbalances tend to impair performance, causing severe performance collapse in minority classes. Conducting effective imbalanced black-box prompt tuning to mitigate the adverse effects of imbalanced data distribution on prompt performance remains a significant challenge. In this paper, we propose black-box prompt tuning with first and zeroth order gradient (BPT-FZG) for handling the imbalanced data. Specifically, BPT-FZG introduces AUC maximization as the objective for prompt tuning and equivalently formulates it as a nonconvex-concave saddle point problem to avoid the construction of sample pairs from opposite classes. Indeed, BPT-FZG optimizes the latent representation of the continuous prompt in the low-dimensional subspace with AUC loss and leverages the first and zeroth order gradients alternately to update the parameters. Furthermore, we establish the theoretical convergence guarantee for BPT-FZG under common assumptions, showing that our method can find a stationary point of the objective function. Our experiments on RoBERTa-large, GPT2-XL, and Llama3 show that BPT-FZG achieves improvement on various imbalanced datasets, emphasizing the effectiveness of our methods.

With the growing significance of network security, the classification of encrypted traffic has emerged as an urgent challenge. Traditional byte-based traffic analysis methods are constrained by the rigid granularity of information and fail to fully exploit the diverse correlations between bytes. To address these limitations, this paper introduces MH-Net, a novel approach for classifying network traffic that leverages multi-view heterogeneous traffic graphs to model the intricate relationships between traffic bytes. The essence of MH-Net lies in aggregating varying numbers of traffic bits into multiple types of traffic units, thereby constructing multi-view traffic graphs with diverse information granularities. By accounting for different types of byte correlations, such as header-payload relationships, MH-Net further endows the traffic graph with heterogeneity, significantly enhancing model performance. Notably, we employ contrastive learning in a multi-task manner to strengthen the robustness of the learned traffic unit representations. Experiments conducted on the ISCX and CIC-IoT datasets for both the packet-level and flow-level traffic classification tasks demonstrate that MH-Net achieves the best overall performance compared to dozens of SOTA methods.

Retrieval-augmented generation (RAG) has revitalized Large Language Models (LLMs) by injecting non-parametric factual knowledge. Compared with long-context LLMs, RAG is considered an effective summarization tool in a more concise and lightweight manner, which can interact with LLMs multiple times using diverse queries to get comprehensive responses. However, the LLM-generated historical responses, which contain potentially insightful information, are largely neglected and discarded by existing approaches, leading to suboptimal results. In this paper, we propose \textit{graph of records} (\textbf{GoR}), which leverages historical responses generated by LLMs to enhance RAG for long-context global summarization. Inspired by the \textit{retrieve-then-generate} paradigm of RAG, we construct a graph by establishing an edge between the retrieved text chunks and the corresponding LLM-generated response. To further uncover the intricate correlations between them, GoR features a \textit{graph neural network} and an elaborately designed \textit{BERTScore}-based objective for self-supervised model training, enabling seamless supervision signal backpropagation between reference summaries and node embeddings. We comprehensively compare GoR with 12 baselines across four long-context summarization datasets, and the results indicate that our proposed method reaches the best performance (\textit{e.g.}, 15%, 8%, and 19% improvement over retrievers w.r.t. Rouge-L, Rouge-1, and Rouge-2 on the WCEP dataset). Extensive experiments further demonstrate the effectiveness of GoR.

Visual Prompt Learning (VPL) has emerged as a powerful strategy for harnessing the capabilities of large-scale pretrained models (PTMs) to tackle specific downstream tasks. However, the opaque nature of PTMs in many real-world applications has led to a growing interest in gradient-free approaches within VPL. A significant challenge with existing black-box VPL methods lies in the high dimensionality of visual prompts, which necessitates considerable API queries for tuning, thereby impacting efficiency. To address this issue, we propose a novel query-efficient framework for blackbox visual prompting, designed to generate input-dependent visual prompts efficiently for large-scale black-box PTMs. Our framework is built upon the insight of reparameterizing prompts using neural networks, improving the typical pretraining-fine-tuning paradigm through the subspace learning strategy to maximize efficiency and adaptability from both the perspective of initial weights and parameter dimensionality. This tuning intrinsically optimizes low-dimensional representations within the well-learned subspace, enabling the efficient adaptation of the network to downstream tasks. Our approach significantly reduces the necessity for substantial API queries to PTMs, presenting an efficient method for leveraging large-scale black-box PTMs in visual prompting tasks. Most experimental results across various benchmarks demonstrate the effectiveness of our method, showcasing substantial reductions in the number of required API queries to PTMs while maintaining or even enhancing performance on downstream tasks.

Large Scale Pre-Trained Language Models (PTMs) have demonstrated unprecedented capabilities across diverse natural language processing tasks. Adapting such models to downstream tasks is computationally intensive and time-consuming, particularly in black-box scenarios common in Language-Model-as-a-Service (LMaaS) environments, where model parameters and gradients are inaccessible. Recently, black-box prompt learning using zeroth-order gradients has emerged as a promising approach to address these challenges by optimizing learnable continuous prompts in embedding spaces, starting with \textit{randomly initialized discrete text prompts}. However, its reliance on randomly initialized discrete prompts limits adaptability to diverse downstream tasks or models. To address this limitation, this paper introduces ZO-PoG, a novel framework that optimizes prompts through a collaborative approach, combining Policy Gradient optimization for initial discrete text prompts and Zeroth-Order optimization for continuous prompts in embedding space. By optimizing collaboratively between discrete and continuous prompts, ZO-PoG maximizes adaptability to downstream tasks, achieving superior results without direct access to the model’s internal structures. Importantly, we establish the sub-linear convergence of ZO-PoG under mild assumptions. The experiments on different datasets demonstrate significant improvements in various tasks compared to the baselines.

Black-box attacks have emerged as a significant threat to deep neural networks. This challenge is particularly difficult in discrete sequential data compared to continuous data. Recently, the Blockwise Bayesian Attack (BBA) leveraging discrete Bayesian optimization with an adapted RBF kernel has gained prominence as a cutting-edge solution. However, it relies solely on alignment information (i.e., positional differences) within the RBF kernel, which may not fully capture the information (such as statistical, structural, and semantic information) inherent in discrete sequential data and potentially lacks the desired inductive bias necessary to approximate the target function accurately. To overcome this limitation, this paper proposes a novel bilevel Bayesian optimization approach to adaptively learn a hybrid space that better captures the similarity between discrete sequences. Specifically, we introduce a multi-kernel mechanism that incorporates multiple types of information, creating a more comprehensive similarity measure. Moreover, we develop a bilevel Bayesian optimization algorithm, where the outer-level objective determines the optimal weights of the multiple kernels, while the inner-level objective identifies the optimal adversarial sequence. Extensive experiments conducted on discrete sequential data demonstrate that our approach ensures secure multi-kernel selection and achieves a higher attack success rate with only a few additional queries, compared to BBA and other traditional optimization strategies.

The rapid emergence of diverse large language models (LLMs) has spurred the development of LLM routers that assign user queries to the most suitable model. However, existing LLM routers typically perform a single-round, one-to-one mapping (\textit{i.e.}, assigning each query to a single model in isolation), which limits their capability to tackle complex tasks that demand the complementary strengths of multiple LLMs. In this paper, we present \textbf{Router-R1}, a reinforcement learning (RL)-based framework that formulates multi-LLM routing and aggregation as a sequential decision process. Router-R1 instantiates the router itself as a capable LLM, leveraging its reasoning ability to interleave "think" actions (internal deliberation) with "route" actions (dynamic model invocation), and integrates each response into its evolving context. To facilitate learning, we employ a lightweight rule-based reward comprising format rewards, final outcome rewards, and a novel cost reward for optimizing the balance between performance and cost, opening a pathway toward enhancing performance-cost trade-offs via RL. Router-R1 also conditions only on simple model descriptors such as pricing, latency, and example performance, enabling strong generalization to unseen model selection. Experiments on seven general and multi-hop QA benchmarks show that Router-R1 outperforms several strong baselines, achieving superior performance while maintaining robust generalization and cost management.

In this paper, we introduce a novel framework for black-box prompt tuning with a subspace learning and selection strategy, leveraging derivative-free optimization algorithms. This approach is crucial for scenarios where user interaction with language models is restricted to API usage, without direct access to their internal structures or gradients, a situation typical in Language-Model-as-a-Service (LMaaS). Our framework focuses on exploring the low-dimensional subspace of continuous prompts. Previous work on black-box prompt tuning necessitates a substantial number of API calls due to the random choice of the subspace. To tackle this problem, we propose to use a simple zeroth-order optimization algorithm to tackle nonconvex optimization challenges with nonsmooth nonconvex regularizers: the Zeroth-Order Mini-Batch Stochastic Proximal Gradient method (ZO-MB-SPG). A key innovation is the incorporation of nonsmooth nonconvex regularizers, including the indicator function of the l0 constraint, which enhances our ability to select optimal subspaces for prompt optimization. The experimental results show that our proposed black-box prompt tuning method on a few labeled samples can attain similar performance to the methods applicable to LMaaS with much fewer API calls.

Temporal Graph Learning, which aims to model the time-evolving nature of graphs, has gained increasing attention and achieved remarkable performance recently. However, in reality, graph structures are often incomplete and noisy, which hinders temporal graph networks (TGNs) from learning informative representations. Graph contrastive learning uses data augmentation to generate plausible variations of existing data and learn robust representations. However, rule-based augmentation approaches may be suboptimal as they lack learnability and fail to leverage rich information from downstream tasks. To address these issues, we propose a Time-aware Graph Structure Learning (TGSL) approach via sequence prediction on temporal graphs, which learns better graph structures for downstream tasks through adding potential temporal edges. In particular, it predicts time-aware context embedding based on previously observed interactions and uses the Gumble-Top-K to select the closest candidate edges to this context embedding. Additionally, several candidate sampling strategies are proposed to ensure both efficiency and diversity. Furthermore, we jointly learn the graph structure and TGNs in an end-to-end manner and perform inference on the refined graph. Extensive experiments on temporal link prediction benchmarks demonstrate that TGSL yields significant gains for the popular TGNs such as TGAT and GraphMixer, and it outperforms other contrastive learning methods on temporal graphs. We release the code at https://github.com/ViktorAxelsen/TGSL.

Encrypted traffic classification is receiving widespread attention from researchers and industrial companies. However, the existing methods only extract flow-level features, failing to handle short flows because of unreliable statistical properties, or treat the header and payload equally, failing to mine the potential correlation between bytes. Therefore, in this paper, we propose a byte-level traffic graph construction approach based on point-wise mutual information (PMI), and a model named Temporal Fusion Encoder using Graph Neural Networks (TFE-GNN) for feature extraction. In particular, we design a dual embedding layer, a GNN-based traffic graph encoder as well as a cross-gated feature fusion mechanism, which can first embed the header and payload bytes separately and then fuses them together to obtain a stronger feature representation. The experimental results on two real datasets demonstrate that TFE-GNN outperforms multiple state-of-the-art methods in fine-grained encrypted traffic classification tasks.

The classification of electrocardiogram (ECG) signals, which takes much time and suffers from a high rate of misjudgment, is recognized as an extremely challenging task for cardiologists. The major difficulty of the ECG signals classification is caused by the long-term sequence dependencies. Most existing approaches for ECG signal classification use Recurrent Neural Network models, e.g., LSTM and GRU, which are unable to extract accurate features for such long sequences. Other approaches utilize 1-Dimensional Convolutional Neural Network (CNN), such as ResNet or its variant, and they can not make good use of the multi-lead information from ECG signals.Based on the above observations, we develop a multi-scale deep residual network for the ECG signal classification task. We are the first to propose to treat the multi-lead signal as a 2-dimensional matrix and combines multi-scale 2-D convolution blocks with 1-D convolution blocks for feature extraction. Our proposed model achieves 99.2% F1-score in the MIT-BIH dataset and 89.4% F1-score in Alibaba dataset and outperforms the state-of-the-art performance by 2% and 3%, respectively, view related code and data at https://github.com/Amadeuszhao/SE-ECGNet.