Singapore

Long Chain-of-Thought (CoT) reasoning enhances large reasoning models&#39; performance but suffers from severe inefficiencies, as models often overthink simple problems or underthink complex ones. Current sequence-level optimizations, like length penalties, are too coarse-grained to distinguish core logic from verbose language, precluding the necessary token-level control for efficient reasoning CoT. To overcome these limitations, we introduce Time-Frequency Token Advantage Clipping (TFAC), a novel training framework designed to build efficient large reasoning models via token-level interventions. Specifically, TFAC functions along two dimensions: 1) The Frequency Dimension: It discourages inefficient loops and encourages deeper exploration by dynamically reducing the advantage scores of high-entropy tokens that are repeatedly generated within a single reasoning path. 2) The Time Dimension: It reduces excessive overthinking of the system by establishing a historical baseline for the occurrence count of each critical token in previously successful trajectories, and clipping the advantages of tokens that exceed this baseline during training. Crucially, to preserve the model&#39;s exploratory capabilities on novel problems, this suppression mechanism is automatically disabled when no historical record of success is available. Experiments conducted on the Deepseek-Distill-32B and Qwen3-8B models show that TFAC outperforms leading baseline methods, improving performance by 2.3 and 3.1 percentage points, respectively, while simultaneously reducing inference costs by 35\% and 28\% in scenarios where correct answers are generated. These results validate the significant efficacy of TFAC in training large reasoning models that are both powerful and highly efficient.

AAAI 2026

Time-Frequency Token Advantage Clipping for Training Efficient Large Reasoning Model

large language models

generation

reinforcement learning

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields 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.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

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 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://2026.aclweb.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/acl2026

Registration Required

ACL 2026

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

Long Chain-of-Thought (CoT) reasoning enhances large reasoning models' performance but suffers from severe inefficiencies, as models often overthink simple problems or underthink complex ones. Current sequence-level optimizations, like length penalties, are too coarse-grained to distinguish core logic from verbose language, precluding the necessary token-level control for efficient reasoning CoT. To overcome these limitations, we introduce Time-Frequency Token Advantage Clipping (TFAC), a novel training framework designed to build efficient large reasoning models via token-level interventions. Specifically, TFAC functions along two dimensions: 1) The Frequency Dimension: It discourages inefficient loops and encourages deeper exploration by dynamically reducing the advantage scores of high-entropy tokens that are repeatedly generated within a single reasoning path. 2) The Time Dimension: It reduces excessive overthinking of the system by establishing a historical baseline for the occurrence count of each critical token in previously successful trajectories, and clipping the advantages of tokens that exceed this baseline during training. Crucially, to preserve the model's exploratory capabilities on novel problems, this suppression mechanism is automatically disabled when no historical record of success is available. Experiments conducted on the Deepseek-Distill-32B and Qwen3-8B models show that TFAC outperforms leading baseline methods, improving performance by 2.3 and 3.1 percentage points, respectively, while simultaneously reducing inference costs by 35\% and 28\% in scenarios where correct answers are generated. These results validate the significant efficacy of TFAC in training large reasoning models that are both powerful and highly efficient.

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.

Please log in

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 

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.

Explanation-guided learning (EGL) has shown promise in aligning model predictions with interpretable reasoning, particularly in computer vision tasks. However, most approaches rely on external annotations or heuristic-based segmentation to supervise model explanations, which can be noisy, imprecise and difficult to scale. In this work, we provide both empirical and theoretical evidence that low-quality supervision signals can degrade model performance rather than improve it. In response, we propose ALIGN, a novel framework that jointly trains a classifier and a masker in an iterative manner. The masker learns to produce soft, task-relevant masks that highlight informative regions, while the classifier is optimized for both prediction accuracy and alignment between its saliency maps and the learned masks. By leveraging high-quality masks as guidance, ALIGN improves both interpretability and generalizability, showing its superiority across various settings. Experiments on the two domain generalization benchmarks, VLCS and Terra Incognita, show that ALIGN consistently outperforms six strong baselines in both in-distribution and out-of-distribution settings. Besides, ALIGN also yields superior explanation quality concerning sufficiency and comprehensiveness, highlighting its effectiveness in producing accurate and interpretable models.

From Attribution to Action: Jointly ALIGNing Predictions and Explanations

Dropout is a widely used regularization technique which improves the generalization ability of a model by randomly dropping neurons. In light of this, we propose Dropout Prompt Learning, which aims for applying dropout to improve the robustness of the vision-language models. Different from the vanilla dropout, we apply dropout on the tokens of the textual and visual branches, where we evaluate the token significance considering both intra-modal context and inter-modal alignment, enabling flexible dropout probabilities for each token. Moreover, to maintain semantic alignment for general knowledge transfer while encouraging the diverse representations that dropout introduces, we further propose residual entropy regularization. Experiments on 11 benchmarks show our method's effectiveness in challenging scenarios like low-shot learning, long-tail classification, and out-of-distribution generalization. Notably, our method surpasses regularization-based methods including KgCoOp by 5.10\% and PromptSRC by 2.13\% in performance on base-to-novel generalization.

Dropout Prompt Learning: Towards Robust and Adaptive Vision-Language Models

Parameter-efficient fine-tuning (PEFT) methods have emerged as a practical solution for adapting large foundation models to downstream tasks, reducing computational and memory costs by updating only a small subset of parameters. Among them, approaches like LoRA aim to strike a balance between efficiency and expressiveness, but often suffer from slow convergence and limited adaptation capacity due to their inherent low-rank constraints. This trade-off hampers the ability of PEFT methods to capture complex patterns needed for diverse tasks. To address these challenges, we propose FRoD, a novel fine-tuning method that combines hierarchical joint decomposition with rotational degrees of freedom. By extracting a globally shared basis across layers and injecting sparse, learnable perturbations into scaling factors for flexible full-rank updates, FRoD enhances expressiveness and efficiency, leading to faster and more robust convergence.
On 20 benchmarks spanning vision, reasoning, and language understanding, FRoD matches full model fine-tuning within 0.09\% accuracy—using only 1.72\% of trainable parameters under identical training budgets.

FRoD: Full-Rank Efficient Fine-Tuning with Rotational Degrees for Fast Convergence

As intelligent systems advance rapidly, human-robot collaboration is becoming increasingly important. Ensuring that the intelligent agent's behaviors match human intentions and value preferences is crucial for effective collaboration, which is termed the value alignment problem. Within the Reinforcement Learning (RL) paradigm, value alignment typically relies on pre-designed reward functions, and Cooperative Inverse Reinforcement Learning (CIRL) is often used to model value alignment as a human-robot game. However, existing works often assume that human is perfectly rational, and can fully obtain robot’s belief on human’s preference. To address this limitation, we propose a Particle Filter-based Hierarchical Dynamic Programming algorithm (PFHDP). By modeling the robot's belief state, this algorithm ensures the correct updates of human's estimate of the robot's belief. This allows human to adopt more targeted pedagogical behaviors to guide the robot based on her understanding of the robot's current belief, achieving belief alignment between human and robot and thereby promoting value alignment more effectively. Furthermore, we run experiments to evaluate the proposed method in two cooperative scenarios against some typical benchmark approaches. The experimental results show that our method can strengthen the alignment of belief states between human and robot, leading to enhanced value alignment.

Belief-Driven Value Alignment for Human-Robot Collaboration

Logical reasoning is a core challenge in natural language understanding and a fundamental capability of artificial intelligence, underpinning scientific discovery, mathematical theorem proving, and complex decision-making. Despite the remarkable progress of large language models (LLMs), most current approaches still rely on forward reasoning paradigms, generating step-by-step rationales from premises to conclusions. However, such methods often suffer from redundant inference paths, hallucinated steps, and semantic drift, resulting in inefficient and unreliable reasoning. In this paper, we propose a novel framework, Hypothesis-driven Backward Logical Reasoning (HBLR). The core idea is to integrate confidence-aware symbolic translation with hypothesis-driven backward reasoning. In the translation phase, only high-confidence spans are converted into logical form, such as first-order logic (FOL), while uncertain content remains in natural language. A translation reflection module further ensures semantic fidelity by evaluating symbolic outputs and reverting lossy ones back to text when necessary. In the reasoning phase, HBLR simulates human deductive thinking by assuming the conclusion is true and recursively verifying its premises. A reasoning reflection module further identifies and corrects flawed inference steps, enhancing logical coherence. Extensive experiments on five reasoning benchmarks demonstrate that HBLR consistently outperforms strong baselines in both accuracy and efficiency. Our code is now available at \url{https://anonymous.4open.science/r/HBLR-AAAI26}

From Hypothesis to Premises: LLM-based Backward Logical Reasoning with Selective Symbolic Translation

Biological foundation models (BioFMs), pretrained on large-scale biological sequences, have recently shown strong potential in providing meaningful representations for diverse downstream bioinformatics tasks. However, such models often rely on millions to billions of training sequences and billions of parameters, resulting in prohibitive computational costs and significant barriers to reproducibility and accessibility—particularly for academic labs. 
To address these challenges, we investigate the feasibility of data pruning for BioFM pretraining and propose a post-hoc influence-guided data pruning framework tailored to biological domains. 
Our approach first introduces a subset-based self-influence formulation that enables efficient estimation of sample importance at low computational cost. Built upon this, we propose two simple yet effective selection strategies: Top-$k$ Influence (Top I) and Coverage-Centric Influence (CCI).
Then, we empirically validate our method on two representative BioFMs: RNA-FM and ESM-C. 
For RNA, our framework consistently outperforms random selection baselines under an extreme pruning rate of over 99\%, which displays our framework's effectiveness.
Furthermore, we demonstrate the generalizability of our framework on protein-related tasks using ESM-C.
In specific, our coreset even outperforms random $10\times$ subsets in both RNA and protein settings, revealing substantial redundancy in biological sequence dataset.
These findings underscore the potential of influence-guided data pruning to substantially reduce the computational cost of BioFM pretraining, paving the way for more efficient, accessible, and sustainable biological AI research.
The code and a technical appendix for better digital viewing are included as supplementary materials and scheduled to be open-sourced upon publication.

Investigating Data Pruning for Pretraining Biological Foundation Models at Scale

Pretrained vision-language models (VLMs), especially CLIP, excel at adapting to downstream tasks through fine-tuning with sufficient high-quality labeled data. 
However, real-world training data often contains noisy labels, leading to significant performance degradation when models are naively fine-tuned on them. 
Existing noisy label learning methods for VLMs typically leverage the model's own pretrained knowledge, either via zero-shot predictions or vanilla self-training based on them, to identify and handle noisy samples. 
Crucially, these approaches blindly trust the VLM's pretrained knowledge, which can introduce endogenous confirmation bias: erroneous pretrained priors lead to incorrect noise detection, further amplifying the bias and corrupting the model.
To overcome this limitation, we propose the Debiased Knowledge Adaptation Framework (DKAF), which empowers the model to challenge and correct potentially flawed zero-shot predictions. 
DKAF operates in three progressive phases:
(1) Clean Sample Selection. We introduce a cross-modal collaborative pseudo-labeling to train a robust noisy label detector, explicitly mitigating confirmation bias by aggregating diverse signals beyond the model's initial zero-shot view.
(2) Noisy Label Refinement. For samples identified as noisy, we apply a dual-modal consistency strategy to selectively correct their labels, leveraging alignment between visual and textual modalities to guide refinement while minimizing reliance on potentially biased internal knowledge.
(3) Model Adaptation. The model is progressively fine-tuned using the jointly curated dataset of selected clean samples and corrected noisy samples, promoting robust adaptation to the target task.
Extensive experiments on nine benchmark datasets (both synthetic and real-world noise) demonstrate that DKAF consistently outperforms state-of-the-art multimodal noisy label learning methods. Notably, under high-noise conditions, DKAF achieves average accuracy improvements of 3.28\%.

Mitigating Endogenous Confirmation Bias in Noisy Label Learning for Vision-Language Models

Self-interpretable models are increasingly valued for their inherent explainability. Among them, part-prototype networks stand out by mimicking human reasoning through the use of learned prototypes. However, their explanations often lack stability, becoming sensitive to subtle input perturbations. In this work, we propose Prototype in Imagery Network (PINet), a framework that improves the stability of prototype-based explanations. Rather than training on all possible input variations, which is computationally infeasible, PINet draws inspiration from visual mental imagery. Specifically, we incorporate empty inputs and apply coarse location guidance to simulate the human ability to imagine rough object features (a process akin to Phantasia). PINet mimics this process by incorporating empty inputs and applying coarse location guidance. These imagined, or uncertain, representations are contrasted with those derived from actual inputs (certain representations). We model the differences between the two by computing similarity at both the feature and prototype levels, allowing uncertainty to be explicitly encoded during prototype learning. Comprehensive evaluations on CUB-200-2011 and Stanford Cars demonstrate that PINet consistently achieves robust accuracy and localization, even under noisy conditions. These results represent the ability of PINet to produce stable and interpretable explanations under uncertainty.

PINet: Improving the Stability of Prototype Networks via Phantasia-Inspired Uncertain Representations

Wikipedia serves as the world's largest and most popular online reference encyclopedia, rich in structured knowledge and authoritative citations. Recently, numerous works have leveraged large language models to automatically generate Wikipedia-like articles. However, existing approaches primarily focus on producing singular narrative-type content, overlooking higher information-density structured elements such as timeline and table.
To address these limitations, we propose WikiMAG, a multi-agent guided framework for generating structured Wikipedia-like articles. This framework employs a collaborative multi-agent mechanism to orchestrate the creation process, featuring three synergistic core components: Progressive planner first constructs the coarse-grained outline framework and then annotate fine-grained types for outline units, encompassing narrative, timeline, and table formats; Reflective inspector dynamically curates high-quality references via multi-round interactive feedback, thereby enhancing the authority and relevance of citations; Versatile writer integrates fine-grained outline details and high-quality reference information to generate information-rich articles, incorporating the three annotated formats. We evaluate WikiMAG on two public datasets, FreshWiki and WikiGenBen, across outline, writing, and verifiability dimensions.
Compared with the best baseline method, our method achieves an average improvement of 6.73 points and 4.39 points in Heading Soft Recall and the METEOR metric (a machine translation and text generation evaluation metric) respectively, and an average increase of 16.84 percentage points in Citation Rate.

WikiMAG: A Multi-Agent Guided Framework for Generating Structured Wikipedia-like Articles

Contrastive Language-Image Pretraining (CLIP) has demonstrated impressive generalization on vision-language tasks by aligning images and short texts. However, its inherent 77-token length constraint limits the capacity to capture complex semantics in long captions. Existing long-text adaptations for CLIP typically rely on either multi-stage training or truncation-based alignment, both inevitably resulting in semantic degradation and cumbersome tuning. Therefore, we propose One-Stage Long-Text Adaptation for CLIP (OneLIP), a unified framework that extends CLIP to understand long captions within a single training stage, eliminating the need for brittle truncation or multi-stage pipelines. OneLIP addresses semantic degradation by introducing two key innovations: (1) Token Refinement and Importance-guided Modeling (TRIM) module, which selects and refines informative tokens via SVD-based contribution scoring and cross-modal relevance modeling; (2) Per-sample Online Hard Negative Mining (PO-HNM) strategy dynamically maintains sample-specific negatives based on dual-consistency difficulty tracking, which is superior to other strategies in long-text scenarios where key semantics are distributed in more scattered positions. Extensive experiments on long-text image retrieval, short-text image retrieval, zero-shot classification, and text-to-image generation demonstrate OneLIP's robustness and versatility across diverse input lengths, offering a semantically faithful solution for long-text adaptation for CLIP.

Downloads

Next from AAAI 2026

From Attribution to Action: Jointly ALIGNing Predictions and Explanations

Similar lecture

Heterogeneous Adaptive Policy Optimization: Tailoring Optimization to Every Token's Nature

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.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 Attribution to Action: Jointly ALIGNing Predictions and Explanations

Similar lecture

Heterogeneous Adaptive Policy Optimization: Tailoring Optimization to Every Token's Nature

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads