Singapore

Transformers are the cornerstone of modern large language models, but their quadratic computational complexity limits efficiency in long-sequence processing. Recent advancements in Mamba, a state space model (SSM) with linear complexity, offer promising efficiency gains but suffer from unstable contextual learning and multitask generalization. Some works conduct layer-level hybrid structures that combine Transformer and Mamba layers, aiming to make full use of both advantages. This paper proposes TransMamba, a novel sequence-level hybrid framework that unifies Transformer and Mamba through shared parameter matrices (QKV and CBx), and thus could dynamically switch between attention and SSM mechanisms at different token lengths and layers. We design the Memory converter to bridge Transformer and Mamba by converting attention outputs into SSM-compatible states, ensuring seamless information flow at TransPoints where the transformation happens. The TransPoint scheduling is also thoroughly explored for balancing effectiveness and efficiency. We conducted extensive experiments demonstrating that TransMamba achieves superior training efficiency and performance compared to single and hybrid baselines, and validated the deeper consistency between Transformer and Mamba paradigms at sequence level, offering a scalable solution for next-generation language modeling.

AAAI 2026

TransMamba: A Sequence-Level Hybrid Transformer-Mamba Language Model

nlp: (large) language models

nlp: generation

ml: deep neural architectures and foundation models

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

ACL 2025

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

poster

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of tasks, yet they generally lack self-awareness—often displaying overconfidence when confronted with questions beyond their knowledge boundaries. This limitation severely hinders their trustworthiness in high-stakes scenarios. Existing calibration methods typically rely on sampling accuracy—derived from multiple outputs—as a proxy for model confidence. However, this coarse-grained metric fails to capture the model’s internal cognitive states, such as confusion, hallucination, or persistent belief in false knowledge.
To address this, we propose \texttt{CogConf} (Cognitive Confidence), a cognitively grounded uncertainty signal that extends sampling accuracy by incorporating the semantic diversity of incorrect answers and the model’s abstention behaviors. By shifting the focus from sampling-based to cognition-oriented uncertainty modeling, \texttt{CogConf} offers a more faithful reflection of the model's internal beliefs. Building on this signal, we introduce \textsc{CogAlign}, a simple yet effective alignment framework that explicitly aligns the model’s verbalized confidence with \texttt{CogConf}, thereby producing uncertainty estimates that better reflect the model’s internal cognition.
Experimental results on six knowledge-intensive in-domain and out-of-domain QA datasets demonstrate that \texttt{CogConf} robustly characterizes the model's internal uncertainty. Building on this foundation, \textsc{CogAlign} guides the model's expression to significantly enhance the trustworthiness and utility of its uncertainty calibration without compromising its underlying QA capabilities, while also demonstrating strong cross-task generalization and output stability. Offering a new pathway toward building more trustworthy LLMs.

From Sampling to Cognition: Modeling Internal Cognitive Confidence in Language Models for Robust Uncertainty Calibration

With the proliferation of distributed data sources, Federated Learning (FL) has emerged as a key approach to enable collaborative intelligence through decentralized model training while preserving data privacy. However, conventional FL algorithms often suffer from performance disparities across clients caused by heterogeneous data distributions and unequal participation, which leads to unfair outcomes. Specifically, we focus on two core fairness challenges, i.e., representation bias, arising from misaligned client representations, and collaborative bias, stemming from inequitable contribution during aggregation, both of which degrade model performance and generalizability. To mitigate these disparities, we propose CoRe-Fed, a unified optimization framework that bridges collaborative and representation fairness via embedding-level regularization and fairness-aware aggregation. Initially, an alignment-driven mechanism promotes semantic consistency between local and global embeddings to reduce representational divergence. Subsequently, a dynamic reward-penalty-based aggregation strategy adjusts each client’s weight based on participation history and embedding alignment to ensure contribution-aware aggregation. Extensive experiments across diverse models and datasets demonstrate that CoRe-Fed improves both fairness and model performance over state-of-the-art baseline algorithms.

CoRe-Fed: Bridging Collaborative and Representation Fairness via Federated Embedding Distillation

Modern multi-view clustering (MVC) is dominated by two paradigms: multi-view fusion and pseudo-label-guided learning. Pseudo-labeling methods can suffer from confirmation bias; their reliance on a fixed-granularity supervision from an initial clustering can cause learned embeddings to drift from the data's true structure and lose discriminative power. Conversely, fusion methods excel at integrating information but often struggle to robustly differentiate between high-quality and noisy views, which can obscure final cluster boundaries and degrade performance. To address these complementary challenges, we propose GAPS (Granularity-Aware Pseudo Supervision), a novel MVC framework. GAPS introduces a granularity-aware supervision mechanism that generates a full hierarchy of pseudo-labels, enabling the selection of a supervision level that best aligns with the data's intrinsic multi-scale structure. Furthermore, to ensure a high-quality supervisory signal, it incorporates a reliability-aware view selection strategy using a novel Separation-Compactness Index (SCI) to identify and leverage the most informative view for pseudo-label generation. This dual approach ensures the supervisory signal is both structurally adaptive and derived from the most reliable source, leading to highly effective final representations. Extensive experiments on synthetic and real-world datasets demonstrate the effectiveness and superiority of GAPS over other competitors.

Multi-View Clustering with Granularity-Aware Pseudo Supervision

Recent advancements in large language models (LLMs) have demonstrated remarkable text generation capabilities. However, controlling specific attributes of generated text remains challenging without architectural modifications or extensive fine-tuning. Current methods typically toggle a single, basic attribute but struggle with precise multi-attribute control. In scenarios where attribute requirements conflict, existing methods lack coordination mechanisms, causing interference between desired attributes. Furthermore, these methods fail to incorporate iterative optimization processes in the controlled generation pipeline. To address these limitations, we propose Conflict-aware, Composite, and Collaborative Controlled Text Generation (C³TG), a two-phase framework for fine-grained, multi-dimensional text attribute control. During generation, C³TG selectively pairs the LLM with the required attribute classifiers from the 17 available dimensions and employs weighted KL-divergence to adjust token probabilities. The optimization phase then leverages an energy function combining classifier scores and penalty terms to resolve attribute conflicts through iterative feedback, enabling precise control over multiple dimensions simultaneously while preserving natural text flow. Experiments show that C³TG significantly outperforms baselines across multiple metrics including attribute accuracy, linguistic fluency, and output diversity, while simultaneously reducing toxicity. These results establish C³TG as an effective and flexible solution for multi-dimensional text attribute control that requires no costly model modifications.

C³TG: Conflict-aware, Composite, and Collaborative Controlled Text Generation

Radiology report generation from chest X-rays is an important task in artificial intelligence with the potential to greatly reduce radiologists' workload and shorten patient wait times. Despite recent advances, existing approaches often lack sufficient disease-awareness in visual representation and adequate vision-language alignment for the specialized requirements of medical image analysis. As a results, these models often overlook critical pathological features in chest X-rays and struggle to generate clinically accurate reports. To address these limitations, we propose a novel dual-stage disease-aware framework for chest X-ray report generation. In Stage 1, our model learns Disease-Aware Semantic Tokens (DASTs) corresponding to specific pathology categories through cross-attention mechanisms and multi-label classification, while simultaneously aligning vision and language representations via contrastive learning. In Stage 2, we introduce a Disease-Visual Attention Fusion (DVAF) module to integrate disease-aware representations with visual features, along with a Dual-Modal Similarity Retrieval (DMSR) mechanism that combines visual and disease-specific similarities to retrieve relevant exemplars, providing contextual guidance during report generation. Extensive experiments on benchmark datasets (i.e., CheXpert Plus, IU X-ray, and MIMIC-CXR) demonstrate that our disease-aware framework achieves state-of-the-art performance in chest X-ray report generation, with significant improvements in clinical accuracy and linguistic quality.

A Disease-Aware Dual-Stage Framework for Chest X-ray Report Generation

Evolutionary algorithms are widely used for solving multi-objective optimization problems. A prominent example is NSGA-III, which is particularly well suited for problems involving more than three objectives, distinguishing it from the classical NSGA-II. Despite its empirical success, the theoretical understanding of NSGA III remains very limited, especially with respect to runtime analysis. A central open problem concerns its population dynamics, which involve controlling the maximum number of individuals sharing the same fitness value during the exploration process. In this paper, we make a significant step towards such an understanding by proving tight runtime bounds for NSGA-III on the bi-objective OneMinMax ($2$-OMM) problem. Firstly, we prove that NSGA-III requires $\Omega(n^2 \log(n) / \mu)$ generations in expectation to optimize $2$-OMM assuming the population size $\mu$ satisfies $n+1 \leq \mu =O(\log(n)^c(n+1))$ where $n$ denotes the problem size and $c<1$ is a constant. Apart from~\cite{opris2025multimodal}, this is the first proven lower runtime bound for NSGA-III on a classical benchmark problem. Complementing this, we secondly improve the best known upper bound of NSGA-III on the $m$-objective OneMinMax problem ($m$-OMM) of $O(n \log(n))$ generations by a factor of $\mu /(2n/m + 1)^{m/2}$ for a constant number $m$ of objectives and population size $(2n/m + 1)^{m/2} \leq \mu \in O(\sqrt{\log(n)} (2n/m + 1)^{m/2})$. This yields tight runtime bounds in the case $m = 2$ and the surprising result that NSGA-III beats NSGA-II by a factor of $\mu/n$ in the expected runtime.

Towards a Rigorous Understanding of the Population Dynamics of the NSGA-III: Tight Runtime Bounds

Multimodal Large Language Models (MLLMs) employing the Mixture-of-Experts (MoE) structure exhibit encouraging results in visual language tasks. However, they struggle with catastrophic forgetting due to a lack of effective collaboration among experts and negative transfer across tasks. This happens because the router typically employed in MoE for managing expert assignments is inadequate when there are significant shifts in data distribution across various tasks. A drop in the effectiveness of earlier tasks is caused by negative transfer, which occurs due to conflicts in shared knowledge between tasks, disturbing the knowledge already acquired. To address these issues, we propose the Knowledge Space Synergy framework in Mixture of Experts (KSS-MoE) for Continual Visual Instruction Tuning (CVIT). It dynamically combines the knowledge subspaces of experts to improve the integration of fine-grained complementary knowledge and collaborative abilities of experts, thus addressing the limitations of the basic router. Furthermore, we introduce a general expert that maintains orthogonal subspaces for shared knowledge, enabling effective cross-task knowledge utilization while reducing negative transfer. Extensive experiments conducted on eight CVIT tasks confirm the excellence of KSS-MoE, showcasing its top-tier performance. Our code is available in the appendix.

KSS-MoE: Knowledge Space Synergy Framework in Mixture of Experts for Continual Visual Instruction Tuning

Multimodal sarcasm detection is a complex task that requires distinguishing subtle complementary signals across modalities while filtering out irrelevant information. Many advanced methods rely on learning shortcuts from datasets rather than extracting intended sarcasm-related features. However, our experiments show that shortcut learning impairs the model's generalization in real-world scenarios. Furthermore, we reveal the weaknesses of current modality fusion strategies for multimodal sarcasm detection through systematic experiments, highlighting the necessity of focusing on effective modality fusion for complex emotion recognition. To address these challenges, we construct MUStARD++$^{R}$ by removing shortcut signals from MUStARD++. Then, a Multimodal Conditional Information Bottleneck (MCIB) model is introduced to enable efficient multimodal fusion for sarcasm detection. Experimental results show that the MCIB achieves the best performance without relying on shortcut learning. The code is provided in the appendix.

Conditional Information Bottleneck for Multimodal Fusion: Overcoming Shortcut Learning in Sarcasm Detection

Evolutionary Algorithms (EAs) have become the most popular tool for solving widely-existed multi-objective optimization problems. In Multi-Objective EAs (MOEAs), there is increasing interest in using an archive to store non-dominated solutions generated during the search. This approach can 1) mitigate the effects of population oscillation, a common issue in many MOEAs, and 2) allow for the use of smaller, more practical population sizes. In this paper, we analytically show that the archive can even further help MOEAs through reusing its solutions during the process of new solution generation. We first prove that using a small population size alongside an archive (without incorporating archived solutions in the generation process) may fail on certain problems, as the population may remove previously discovered but promising solutions. We then prove that reusing archive solutions can overcome this limitation, resulting in at least a polynomial speedup on the expected running time. Our analysis focuses on the well-established SMS-EMOA algorithm applied to the commonly studied OneJumpZeroJump problem as well as one of its variants. We also show that reusing archive solutions can be better than using a large population size directly. Finally, we show that our theoretical findings can generally hold in practice by experiments on four well-known practical optimization problems -- multi-objective 0-1 Knapsack, TSP, QAP and NK-landscape problems -- with realistic settings.

Not Just for Archiving: Provable Benefits of Reusing the Archive in Evolutionary Multi-objective Optimization

Mixture of experts (MoE) dynamically routes inputs to specialised expert networks to scale model capacity with low inference overhead. However, the excessive parameters growth in MoE models poses challenges in low-resource settings. To address these issues, MoE with parameter-efficient fine-tuning (PEFT) methods have emerged as a lightweight adaptation paradigm that distributes knowledge among experts via multiple LoRA blocks. 
Existing MoE-PEFT methods can be broadly categorized into External and Internal PEFT methods. 
External PEFT methods incorporate lightweight models into existing MoE architectures without modifying their routing, which limits the model’s parameters efficiency. 
To overcome these issues, Internal PEFT methods integrate MoE architectures into PEFT, enabling minimal parameters overhead. However, they still face two major challenges: (1) lack of expert functional differentiation, resulting in overlapping specialization across modules, and (2) absence of a structured attribution mechanism to guide expert selection based on semantic relevance.
To alleviate these challenges, we propose TopicLoRA, a novel three-stage framework that leverages topic knowledge as semantic anchors to guide expert allocation. Specifically, (1) to address expert redundancy, we construct a topic-level prior graph using Graph Neural Network-enhanced representation learning over Big-bench categories, enforcing structural separation among expert embeddings, and (2) to introduce semantic attribution, we design a dual-loss training mechanism that softly aligns input-query relevance with topic-guided routing distributions via KL divergence.
Extensive experiments on representative datasets (e.g., MMLU, GSM8K, Flanv2) demonstrate that TopicLoRA outperforms state-of-the-art PEFT baselines by 2.40\% on average in accuracy. Notably, the maximum improvement is 4.21\%. Furthermore, ablation studies demonstrate that our framework's robustness to intricate topics and input sequence variations, which stems from the dual-loss training mechanism. Code is available at https://anonymous.4open.science/r/TopicLoRA-00AF/.

Downloads

Next from AAAI 2026

From Sampling to Cognition: Modeling Internal Cognitive Confidence in Language Models for Robust Uncertainty Calibration

Similar lecture

Mamba Knockout for Unraveling Factual Information Flow

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2026

From Sampling to Cognition: Modeling Internal Cognitive Confidence in Language Models for Robust Uncertainty Calibration

Similar lecture

Mamba Knockout for Unraveling Factual Information Flow

Downloads