Xiaoquan Yi, Haixing Wu, Haozhao Wang, Yichen Li, Yuhua Li, Rui Zhang, Ruixuan Li. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

By allowing each client to refer to the knowledge from other clients while retaining their specific characteristics, partial knowledge sharing has become one of the main approaches to realizing personalized federated learning (pFL). Representative techniques of partial knowledge sharing propose sharing the feature extractor while customizing the classifier head of the neural network. Although such methods achieve great success, the underlying principle behind them remains yet to be comprehensively understood. A fundamental problem is whether it is really appropriate to fully share the feature extractor. Based on the theory of neural collapse, in this paper, we demonstrate both theoretically and empirically that the feature extractor should be partially shared rather than fully shared. More specifically, we identify a substantial inconsistency between the fused global feature representations and expected local feature representations, and thus it is necessary to preserve partially customized layers of the feature extractor for enhancing personalized representations. Based on this discovery, we further propose a novel method called pFedDKS which detaches the shared global knowledge and customized local knowledge by providing detached feature prototypes. Extensive experiments on various datasets and models show that pFedDKS outperforms state-of-the-arts.

With the increasing sophistication of Large Language Models (LLMs), it is crucial to develop reliable methods to accurately identify whether an interlocutor in real-time dialogue is human or chatbot. However, existing detection methods are primarily designed for analyzing full documents, not the unique dynamics and characteristics of dialogue. These approaches frequently overlook the nuances of interaction that are essential in conversational contexts. This work identifies two key patterns in dialogues: (1) Human-Human (H-H) interactions exhibit significant bidirectional sentiment influence, while (2) Human-Chatbot (H-C) interactions display a clear asymmetric pattern. We propose an innovative approach named ChatbotID, which applies the Granger Causality Test (GCT) to extract a novel set of interactional features that capture the evolving, predictive relationships between conversational attributes. By synergistically fusing these GCT-based interactional features with contextual embeddings, and optimizing the model through a meticulous loss function. Experimental results across multiple datasets and detection models demonstrate the effectiveness of our framework, with significant improvements in accuracy for distinguishing between H-H and H-C dialogues.

Deep learning models, usually trained on large datasets, have witnessed great achievements for both classification and regression tasks during past years. However, in practical settings, the datasets are commonly imbalanced where the number of data samples differs among labels (i.e.,categories or target values), leading to serious performance degradation of the trained models. Although many prior works have been proposed to solve the data imbalance problem for the classification task, there are still limited works considering the regression task. In this work, we identify the distinct challenges in regression problems compared to traditional classification problems, such as managing the continuous nature of the target variable and navigating fuzzy decision-making boundaries. To address these challenges, we propose an Adaptive Weighted Statistical Space Smoothing method (AWSSS), which alleviates the impact of imbalanced data by learning from continuously valued, imbalanced data and extending local similarity across the entire target range. AWSSS extracts statistical values from the original dataset and smoothes statistical features. Subsequently, it aligns features across different regions and applies adaptive weights to facilitate knowledge transfer between adjacent areas. Finally, AWSSS calculates the loss values, which are used to refine the model’s parameters through iterative updates. Extensive experiments conducted on various datasets demonstrate the effectiveness of the proposed method as compared to state-of-the-art methods.

Federated Learning (FL) has emerged as a foundational paradigm that enables collaborative training of deep neural networks across distributed clients while ensuring data privacy. However, most of the existing research primarily concentrates on supervised federated learning tailored for specific downstream tasks. In this paper, we concentrate on unsupervised federated learning, where clients collaborate to identify data patterns, structures, or representations. Contrastive learning methods have proven highly effective for unsupervised learning in server-based environments. However, a straightforward adaptation of the contrastive learning approach for this setting falls short. Upon analyzing client behaviors during training, we identified an unstable training process stemming from NonIID issues, which can result in diminished performance. To address this problem, we propose a novel Temperature-Adaption based unsupervised Federated prototype learning approach, termed FedTA. This method incorporates two straightforward yet potent solutions: 1. Prototype Enhancement and 2. Temperature-Adaption. Experimental results show that our proposed approach surpasses the state-of-the-art, achieving a performance accuracy improvement of up to 5.3%.

Federated Learning (FL) allows massive clients to collaboratively train a global model without revealing their private data. Because of the participants’ not independently and identically distributed (non-IID) statistical characteristics, it will cause divergence among the client’s Deep Neural Network model weights and require more communication rounds before training can be converged. Moreover, models trained from non-IID data may also extract biased features and the rationale behind the model is still not fully analyzed and exploited. In this paper, we propose eXplainable-Fed (XFed) which is a novel client selection mechanism that takes both accuracy and explainability into account. Specifically, XFed selects participants in each round based on a small test set’s accuracy via cross-entropy loss and interpretability via XAI-accuracy. XAI-accuracy is calculated by Intersection over Union Ratio between the heat map and the truth mask to evaluate the overall rationale of accuracy. The results of our experiments show that our method has comparable accuracy to state-of-the-art methods specially designed for accuracy while increasing explainability by 14%-35% in terms of rationality.

In this paper, we propose an efficient and practical algorithm to dynamically adapt the Lagrange multipliers for each macroblock based on the context of the neighboring or upper layer blocks to improve rate-distortion performance. Our method improves the accuracy for the detection of true motion vectors as well as the most efficient encoding modes for luma, which are used for deriving the motion vectors, and modes for chroma. Simulation results for H.264/advanced video coding video demonstrate that our method reduces bit rate significantly and achieves peak signal-to-noise ratio gain over those of the joint model (JM) software for all sequences tested, with negligible extra computational cost. The improvement is particularly significant for high motion high-resolution videos. This paper describes our work that led to our Joint Video Team adopted contribution (included in software JM 12.0 onward), collectively known as context adaptive Lagrange multiplier (CALM).

The increasing interest in higher fidelity video initiated the need of higher efficient coding. Color pictures are usually compressed in a luma-chroma coordinate space. One of the important components in H.264 encoding is to find optimal motion vectors and modes to minimize both luma and chroma coefficients bits. This paper proposes a new simple and efficient method to adjust Lagrange multipliers based on the context (context adaptive Lagrange multiplier or CALM), which improves the accuracy for the detection of true motion vectors as well as the most efficient encoding modes for luma which are used for deriving the motion vectors and modes for chroma. Simulation results show that the chroma bit rates can be reduced by 4.36% and 4.80% for U and V components respectively when compared with that of the JM reference software (version 10.2). In addition, the coding efficiency improvement is comparable to the more complicated rate-distortion optimized (RDO) mode decision techniques.

Motion estimation is a critical yet computationally intensive task for video encoding. In this paper, we present an enhancement over a normalized partial distortion search (NPDS) algorithm to further reduce block matching motion estimation complexity while retaining video fidelity. The novelty of our algorithm is that, in addition to the halfway-stop technique in NPDS, a dual-halfway-stop (DHS) method, which is based on a dynamic threshold, is proposed, so that block matching is not performed against all matching candidates. An adaptive search range (ASR) mechanism based on inter block distortion further constrains the searching process. Simulation results show that the proposed algorithm has a remarkable computational speedup when compared to that of full search and NPDS algorithms. Particularly, it requires less computation by 92-99% and encounters an average of only 0.08 dB PSNR video degradation when compared to that of full search. The speedup is also very significant when compared to that of fast motion estimation algorithms. This paper describes our work that led to our joint video team (JVT) adopted contribution (included in software JM 10.1 onwards) as well as later enhancements, collectively known as simplified and unified multi-hexagon search (SUMH), a simplified fast motion estimation.

Block motion estimation is a critical, yet computationally intensive, task for video encoding. Many fast block matching algorithms have been developed. Partial distortion search (PDS) algorithms generally produce less video quality degradation of the predicted images than those of conventional fast block matching algorithms (BMAs). However, the speedup gain of PDS algorithms is usually limited. We present an enhancement over a normalized PDS (NPDS) algorithm to reduce block matching motion estimation complexity further and improve video fidelity. The novelty of our algorithm is that, in addition to the halfway-stop technique in NPDS, a dual-halfway-stop (DHS) method, which is based on a dynamic threshold, is proposed so that block matching is not performed against all searching points. The dynamic threshold is obtained via a linear model utilizing already computed distortion statistics. An adaptive search range mechanism based on inter block distortion further constrains the searching process. Simulation results show that the proposed algorithm has a remarkable computational speedup. Particularly, it requires 92.0-99.4% less computation than full search (FS) and 8.0-91.0% less than NPDS. It encounters an average of 0.07 dB video degradation in PSNR performance compared to FS whereas it gains 0.01-0.12 dB over the NPDS algorithm.

In order to achieve constant quality across the whole video sequence under the channel bandwidth and buffer constraints, it is necessary to allocate more bits to frames with scene changes or high complexity and fewer bits to low complexity frames. In this work, we propose a new frame layer bit allocation scheme for H.264 video coding using mean absolute difference (MAD) ratio, which is the ratio of MAD of current frame to the average MAD from the starting frame up to the previous frame in a GOP. We provide a theoretical justification of MAD ratio as a measure of frame complexity. Bit budget is allocated to frames according to their MAD ratios, combined with the bits computed based on their buffer status. Simulation results show that the H.264 coder, using our proposed algorithm with virtually little computational complexity added, effectively alleviates visual quality degradation caused by high motion or scene changes. Our proposed algorithm significantly reduces the standard deviation of PSNR, hence producing a nearly constant video quality throughout the whole video sequence, when compared with other existing video schemes.