Guodong Du, Zitao Fang, Jing Li, Junlin Li, Runhua Jiang, Shuyang Yu, Yifei Guo, Yangneng Chen, Sim Kuan Goh, Ho-Kin Tang, Daojing He, Honghai Liu, Min Zhang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Junlin Li, Guodong Du, Jing Li, Sim Kuan Goh, Wenya Wang, Yequan Wang, Fangming Liu, Ho-Kin Tang, Saleh Alharbi, Daojing He, Min Zhang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Fine-tuning pre-trained models on targeted datasets enhances task-specific performance but often comes at the expense of generalization.Model merging techniques, which integrate multiple fine-tuned models into a single multi-task model through task arithmetic, offer a promising solution.However, task interference remains a fundamental challenge, leading to performance degradation and suboptimal merged models.Existing approaches largely overlooked the fundamental roles of neurons, their connectivity, and activation, resulting in a merging process and a merged model that does not consider how neurons relay and process information.In this work, we present the first study that relies on neuronal mechanisms for model merging.Specifically, we decomposed task-specific representations into two complementary neuronal subspaces that regulate input sensitivity and task adaptability.Leveraging this decomposition, we introduced NeuroMerging, a novel merging framework developed to mitigate task interference within neuronal subspaces, enabling training-free model fusion across diverse tasks.Through extensive experiments, we demonstrated that NeuroMerging achieved superior performance compared to existing methods on multi-task benchmarks across both natural language and vision domains.Our findings highlighted the importance of aligning neuronal mechanisms in model merging, offering new insights into mitigating task interference and improving knowledge fusion.Our project is available at https://ZzzitaoFang. github.io/projects/NeuroMerging/.

Electroencephalography (EEG) has wide-ranging applications, from clinical diagnosis to brain-computer interfaces (BCIs). With the increasing volume and variety of EEG data, there has been growing interest in establishing foundation models (FMs) to scale up and generalize neural decoding. Despite showing early potential, applying FMs to EEG remains challenging due to substantial inter-subject, inter-task, and inter-condition variability, as well as diverse electrode configurations across recording setups. To tackle these open challenges, we propose **NeurIPT**, a foundation model tailored for diverse EEG-based **Neur**al **I**nterfaces with a **P**re-trained **T**ransformer by capturing both homogeneous and heterogeneous spatio-temporal characteristics inherent in EEG signals. Temporally, we introduce Amplitude-Aware Masked Pretraining (AAMP), masking based on signal amplitude rather than random intervals, to learn robust representations across varying signal intensities beyond local interpolation. Moreover, this temporal representation is enhanced by a progressive Mixture-of-Experts (MoE) architecture, where specialized expert subnetworks are progressively introduced at deeper layers, adapting effectively to the diverse temporal characteristics of EEG signals. Spatially, NeurIPT leverages the 3D physical coordinates of electrodes, enabling effective transfer across varying EEG settings, and develops Intra-Inter Lobe Pooling (IILP) during fine-tuning to efficiently exploit regional brain features. Empirical evaluations across nine downstream BCI datasets, via fine-tuning and training from scratch, demonstrated NeurIPT consistently achieved state-of-the-art performance, highlighting its broad applicability and robust generalization. Our work pushes forward the state of FMs in EEG and offers insights into scalable and generalizable neural information processing systems.

Transforming Baimiao sketches into full-color Chinese brush paintings is significant in the context of rising market demands and exploring new possibilities in the realm of AI art creation. While several pixel-based works demonstrated the capability of generative adversarial network (GAN) for colorization, several challenges remain, e.g., the inappropriate thickness of lines in sketches and the shortage of local detail. To address the challenges, this paper introduces the double-scale discriminator GAN (DSD-GAN), with double discriminators that target realistic image generation at two scales. The generator of DSD-GAN integrates a Unet-based architecture with a convolutional attention module capable of focusing on critical areas to enhance the painting color and detail accuracy. Moreover, an additional enhancement block, inspired by the perceptual field model, is adopted to improve the precision of local details in the generated artwork. Extensive verification on a diverse dataset of Chinese brush paintings, including two artistic subject matter still life and landscape, demonstrates DSD-GAN’s superiority over state-of-the-art Image-to-Image translation methods across various metrics, including PSNR, SSIM, MSE, and PI. Qualitative visual assessments further confirm its effectiveness. The source code is accessible on GitHub upon paper acceptance.

Spiking neural networks (SNNs) have gained prominence for their potential in neuromorphic computing and energy-efficient artificial intelligence, yet optimizing them remains a formidable challenge for gradient-based methods due to their discrete, spike-based computation. This paper attempts to tackle the challenges by introducing Cosine Annealing Differential Evolution (CADE), designed to modulate the mutation factor (F) and crossover rate (CR) of differential evolution (DE) for the SNN model, i.e., Spiking Element Wise (SEW) ResNet. Extensive empirical evaluations were conducted to analyze CADE. CADE showed a balance in exploring and exploiting the search space, resulting in accelerated convergence and improved accuracy compared to existing gradient-based and DE-based methods. Moreover, an initialization method based on a transfer learning setting was developed, pretraining on a source dataset (i.e., CIFAR-10) and fine-tuning the target dataset (i.e., CIFAR-100), to improve population diversity. It was found to further enhance CADE for SNN. Remarkably, CADE elevates the performance of the highest accuracy SEW model by an additional 0.52 percentage points, underscoring its effectiveness in fine-tuning and enhancing SNNs. These findings emphasize the pivotal role of a scheduler for F and CR adjustment, especially for DE-based SNN. Source Code on Github: https://github.com/Tank-Jiang/CADE4SNN.

While fine-tuning pretrained models has become common practice, these models often underperform outside their specific domains. Recently developed model merging techniques enable the direct integration of multiple models, each fine-tuned for distinct tasks, into a single model. This strategy promotes multitasking capabilities without requiring retraining on the original datasets. However, existing methods fall short in addressing potential conflicts and complex correlations between tasks, especially in parameter-level adjustments, posing a challenge in effectively balancing parameter competition across various tasks. This paper introduces an innovative technique named **PCB-Merging** (Parameter Competition Balancing), a *lightweight* and *training-free* technique that adjusts the coefficients of each parameter for effective model merging. PCB-Merging employs intra-balancing to gauge parameter significance within individual tasks and inter-balancing to assess parameter similarities across different tasks. Parameters with low importance scores are dropped, and the remaining ones are rescaled to form the final merged model. We assessed our approach in diverse merging scenarios, including cross-task, cross-domain, and cross-training configurations, as well as out-of-domain generalization. The experimental results reveal that our approach achieves substantial performance enhancements across multiple modalities, domains, model sizes, number of tasks, fine-tuning forms, and large language models, outperforming existing model merging methods.

Darwinian evolution of the biological brain is documented through multiple lines of evidence, although the modes of evolutionary changes remain unclear. Drawing inspiration from the evolved neural systems (e.g., visual cortex), deep learning models have demonstrated superior performance in visual tasks, among others. While the success of training deep neural networks has been relying on back-propagation (BP) and its variants to learn representations from data, BP does not incorporate the evolutionary processes that govern biological neural systems. This work proposes a neural network optimization framework based on evolutionary theory. Specifically, BP-trained deep neural networks for visual recognition tasks obtained from the ending epochs are considered the primordial ancestors (initial population). Subsequently, the population evolved with differential evolution. Extensive experiments are carried out to examine the relationships between Darwinian evolution and neural network optimization, including the correspondence between datasets, environment, models, and living species. The empirical results show that the proposed framework has positive impacts on the network, with reduced over-fitting and an order of magnitude lower time complexity compared to BP. Moreover, the experiments show that the proposed framework performs well on deep neural networks and big datasets.

Neural architecture search (NAS) automates neural network design by using optimization algorithms to navigate architecture spaces, reducing the burden of manual architecture design. While NAS has achieved success, applying it to emerging domains, such as analyzing unstructured 3D point clouds, remains underexplored due to the data lying in non-Euclidean spaces, unlike images. This paper presents Success-History-based Self-adaptive Differential Evolution with a Joint Point Interaction Dimension Search (SHSADE-PIDS), an evolution-ary NAS framework that encodes discrete deep neural network architectures to continuous spaces and performs searches in the continuous spaces for efficient point cloud neural architectures. Comprehensive experiments on challenging 3D segmentation and classification benchmarks demonstrate SHSADE-PIDS's capabilities. It discovered highly efficient architectures with higher accuracy, significantly advancing prior NAS techniques. For segmentation on SemanticKITTI, SHSADE-PIDS attained 64.51% mean IoU using only 0.55M parameters and 4.5GMACs, reducing overhead by over 22-26X versus other top methods. For ModelNet40 classification, it achieved 93.4% accuracy with just 1.31M parameters, surpassing larger models. SHSADE-PIDS provided valuable insights into bridging evolutionary algorithms with neural architecture optimization, particularly for emerging frontiers like point cloud learning.

Brain computer interface (BCI) technique has been demonstrated that human intentions or stimulus perception can be recognized using EEG signal recorded from the human scalp. When an intention is initiated in the brain or an external stimulus is perceived, the underlying relevant processing alters brain activity. This alteration in brain activity can be reflected in EEG signal. The intention or stimulus perception is therefore classified based on the alteration in brain activity. It might be difficult to classify brain activities in the perception of shape-analogous English letters because the similar shape could lead to less difference in brain activity. In order to explore classification feasibility and classification performance of shape-analogous letters using EEG signal, we performed an experiment of shape-analogous letter perception, in where participants perceived four letters (i.e., 'p', 'q', 'b' and 'd') while EEG signal was recorded. The F-score method was employed to assess the discriminative power for each feature, and a subgroup of features with high discriminative powers was then selected and fed into classifiers. Five classifiers (i.e., k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Random Forest (RF) and AdaBoost (ADA)), which are either pervasive or advanced in the field of machine learning, were utilized to classify brain activities in perception of shape-analogous letters. For each classifier, its parameters and the number of used features were optimized. Based on the performance comparison among the classifiers, Random Forest (RF) classifier achieved a maximal accuracy of 74.1%, but it was not statistically significantly better than the SVM. Our study demonstrated that brain activities in perception of shape-analogous English letters can be classified based on EEG signal and showed that random forest classifier outperformed other classifiers according to the results of comparison.

Neural information recorded from electroencephalogram (EEG) provides new possibilities for diagnosis of brain abnormalities, cognitive monitoring, etc. However, many artifacts, such as eye blink and muscle movements, impact and contaminate EEG data. While traditional techniques proposed for artifact removal identified artifact on two-way data, (spatial x temporal), multidimensional nature of EEG data (spatial x temporal x spectral x condition x trial) is overlooked. In this work, we investigate the use of multiway analysis/tensor factorization on the extended EEG tensor (spatial x temporal x spectral), which is constructed from continuous wavelet transform, using Block Term Decomposition (BTD) of rank-(Lr, Lr, 1) for artifact removal. Eight different carefully designed experiments to study artifact typically produced by voluntarily, and sometimes involuntarily, behaviors using a subject were performed and analyzed. After the BTD decomposition, artifacted components are automatically identified removed using spatial and temporal features. The reconstructed signal from proposed method suppresses artifact while retains the signal texture of eight types of artifact investigated.

Challenging multi-modal optimization problems have been very successfully solved by evolutionary computation (EC) techniques. To date, many methods have been proposed on evolutionary optimization for both single and multiobjective large scale problems. In the age of Big Data, there is an urge to take evolutionary optimization techniques to the next level for solving problems with even larger scales: thousands and millions of variables. These problems arise in many domains ranging from bioinformatics, to neuroscience and social simulations. In this paper, we investigate the use of EC to solve Big electroencephalography (EEG) data optimization problems with thousands of variables. The optimization problem attempts to identify maximum information that should be kept from a signal while minimizing the artifact. The high level of epistasis inherent in a signal can slow down the evolution. Therefore, we investigate the advantages of optimizing the problem in the frequency domain with different thresholds as opposed to the time domain. We propose synthetic EEG data sets of various scale and noise level. These data sets were the basis for the Optimization of Big Data 2015 Competition (BigOpt), CEC 2015. Two state-of-art multiobjective evolutionary algorithms (MOEAs) were evaluated. The results of this work suggest that frequency representation of the signals facilitates dimensionality reduction for big scale optimization of time series data, and hence provides faster and better quality solutions for EEG data cleaning. Moreover, the results suggest that existing state-of-art multiobjective evolutionary computation methods are extremely slow. Methods that can optimize the problem faster and with high quality are needed.