United States

Tiny Recursive Models (TRMs) perform iterative reasoning with an Adaptive Computation Time (ACT)-style loop, but their supervised training targets can be brittle, and their halting behavior can be difficult to tune. We introduce TRM-Planner, a two-stage teacher-cache distillation recipe that shifts compute to an offline teacher-cache stage. A frozen TRM checkpoint is unrolled for multiple refinement steps and stochastic rollouts; for each instance, we cache a small set of teacher entries (tokens, logits, step index, and quality metadata). A student TRM is then trained with the standard TRM objective plus a distillation loss computed from cached entries. Across Sudoku-Extreme and ARC-AGI-1/2, TRM-Planner shows an improvement over our reproduced TRM baseline while leaving student-time inference unchanged. On ARC1/ARC2 with 7M parameters, the two-attempt accuracy (pass@2) increases from 43.1% to 48.1% and 6.7% to 9.2%, respectively.

ACL 2026

 TRM-Planner: Offline Target Planning and Distillation for Tiny Recursive Models

Congratulations to all the researchers who have qualified for the AMA Research Challenge virtual poster symposium and semifinals!

Welcome to the **largest national, multi-specialty research event** featuring research from nearly 1,000 medical students, residents, fellows, and international medical graduates across six topic areas - basic science, clinical and translational research, clinical vignettes, health systems science, medical education, and public health and health policy.

#### Access the Event
Use the buttons below or navigation menu to access the event. 

[![](https://assets.underline.io/markdown_image/1/image/d4076f2146939317a29eb9f69c53ad41.png)](https://underline.io/events/506/sessions?searchGroup=all)
[![](https://assets.underline.io/markdown_image/1/image/6bcee0b523a4874c4292ca0415d88dde.png)](https://underline.io/events/506/posters?searchGroup=all)

#### Event Details
The 2025 AMA Research Challenge includes the following symposium halls:
* **Semifinals hall:** Features research from 54 medical students, residents and international medical graduates that received top scores during first round scoring. 
* **Poster symposium hall:** Features research from nearly 1,000 medical students, residents and international medical graduates categorized by topics and subtopics.

**Here’s how to participate:**
* **View posters and presentations** on a variety of topics and specialties
* **Score posters in the semifinals** to help decide the top five finalists (AMA member exclusive opportunity)
* **Connect with presenters** and provide feedback to help advance their research
* **New this year: Watch live poster presentations** by topic area and engage directly with researchers

At the conclusion of this event, the AMA announced the top five finalists who advanced to the AMA Research Challenge final round, presented their research to an elite panel of judges, and competed to win a $10,000 grand prize presented by Laurel Road.

**Addison Shenk, a fourth-year medical student from Edward Via Virginia College of Osteopathic Medicine, was the winner of the 2025 AMA Research Challenge and was awarded the grand prize of $10,000, presented by Laurel Road.**

#### A message from Dr. Bobby Mukkamala, AMA President
<div style="position: relative; width: 100%; padding-bottom: 56.25%; height: 0; overflow: hidden;">
  <iframe 
    src="https://underline.io/embed/136894-ama-video?t=0"
    title="AMA Welcome Video"
    allowfullscreen
    style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border: 0;">
  </iframe>
</div>


#### Recognized Institutions
In 2025, 1,400 abstracts were submitted across six topic areas representing 174 medical schools and more than 150 residency and health care institutions. [See if your institution is represented this year!](https://drive.google.com/file/d/1eFrTqCmCeobx9pr5drJK2hllvsIpFNOY/view?usp=drive_link)

#### Grand Prize Sponsor
AMA would like to thank Laurel Road, a brand of KeyBank, for sponsoring a $10,000 grand prize for the winner of the AMA Research Challenge. Laurel Road and KeyBank are not involved in winner selection process. The winner will be announced during the AMA Research Challenge final round. Payment of the grand prize will be directly from Laurel Road/KeyBank. See the [AMA Research Challenge Official Rules for terms and conditions](https://www.ama-assn.org/system/files/research-challenge-official-rules.pdf).

[![](https://assets.underline.io/uploads/markdown_image/1/image/611e331d4a0eba0920f240cfc04b4d77.png)](https://www.laurelroad.com/)

#### Code of Conduct
<html>
AMA has a code of conduct for its meetings – we ask all attendees to adhere to this code to participate in this year’s meeting. Please read and <a href="https://www.ama-assn.org/about/code-conduct" target="_blank">learn more here</a> about your conduct requirements.
<br><br>
	</html>
	

#### Semifinals Judges

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Need help? Contact us at <support@underline.io>.

Please log in to explore this event.

If you have used the AMA Research Challenge platform previously, please click the ‘Log-in’ button below. If you do not remember your password, please click “forgot password” and follow prompts to reset your password. After resetting your password, please refresh your webpage to access the platform. 

Register to explore research in a variety of topics. AMA members get exclusive access to score posters in the semifinals. 

**Please use [this link here](https://www.ama-assn.org/member-benefits/events/ama-research-challenge) to register for free. **

Registration Is Required

AMA Research Challenge 2025 

Join the AMA Research Challenge poster symposium and semifinals to explore top research posters and score posters in the semifinals (an AMA member exclusive opportunity).

poster

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. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<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>
<br><br><br><br>

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

Registration Required

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

Large Language Models (LLMs) exhibit social biases, which can lead to harmful stereotypes and unfair outcomes. We propose Multi-Persona Thinking (MPT), a simple inference-time framework that reduces social bias by encouraging reasoning from multiple perspectives. MPT guides the model to consider contrasting social identities, such as male and female, together with a neutral viewpoint. These viewpoints then interact through an iterative reasoning process to identify and correct biased judgments. This design transforms the potential weakness of persona assignment into a mechanism to mitigate bias. We evaluate MPT on two widely used bias benchmarks with both open-source and closed-source models. Our results show that MPT achieves a lower bias than the existing prompting-based methods while maintaining the core reasoning ability.

 Multi-Persona Thinking for Bias Mitigation in Large Language Models

The combination of Mixture-of-Experts (MoE) and Low-Rank Adaptation (LoRA) has shown significant potential for enhancing the multi-task learning capabilities of Large Language Models. However, existing methods face two primary challenges: (1)Imprecise Routing in the current MoE-LoRA method fails to explicitly match input semantics with expert capabilities, leading to weak expert specialization. (2)Uniform weight fusion strategies struggle to provide adaptive update strengths, overlooking the varying complexity of different tasks. To address these limitations, we propose \textbf{SAMoRA} (\textbf{S}emantic-\textbf{A}ware \textbf{M}ixture \textbf{o}f Lo\textbf{RA} Experts), a novel parameter-efficient fine-tuning framework tailored for task-adaptive learning. Specifically, A \textbf{Semantic-Aware Router} is proposed to explicitly align textual semantics with the most suitable experts for precise routing. A \textbf{Task-Adaptive Scaling} mechanism is designed to regulate expert contributions based on specific task requirements dynamically. In addition, a novel regularization objective is proposed to jointly promote expert specialization and effective scaling. Extensive experiments on multiple multi-task benchmarks demonstrate that SAMoRA significantly outperforms the state-of-the-art methods and holds excellent task generalization capabilities. Code is available at \url{https://github.com/boyan-code/SAMoRA}

 SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning

Finite-State Transducers (FSTs) are effective models for string-to-string rewriting tasks, often providing the efficiency necessary for high-performance applications, but constructing transducers by hand is difficult. In this work, we propose a novel method for automatically constructing unweighted FSTs following the hidden state geometry learned by a recurrent neural network. We evaluate our methods on real-world datasets for morphological inflection, grapheme-to-phoneme prediction, and historical normalization, showing that the constructed FSTs are highly accurate and robust for many datasets, massively outperforming classical transducer learning algorithms by up to 87% accuracy on held-out test sets.

 Neural Induction of Finite-State Transducers

Although machine unlearning is essential for removing private, harmful, or copyrighted content from LLMs, current benchmarks often fail to faithfully represent the true ``forgetting scope'' learned by the model. 
We formalize two distinct unlearning granularities, domain-level and instance-level, and propose \BiForget, an automated framework for synthesizing high-quality forget sets.
Unlike prior work relying on \emph{external} generators, \BiForget exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting. 
Our experiments across diverse benchmarks show that it achieves a superior balance of relevance, diversity, and efficiency. 
Quantitatively, in the Harry Potter domain, it improves relevance by ${\sim}20$ and diversity by ${\sim}$0.05 while \emph{halving} the total data size compared to SOTAs. 
Ultimately, it facilitates more robust forgetting and better utility preservation, providing a more rigorous foundation for evaluating LLM unlearning.

From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning

Causal self-attention provides positional information to Transformer decoders. Prior work has shown that stacks of causal self-attention layers alone induce a positional bias in attention scores toward earlier tokens. However, this differs from the bias toward later tokens typically observed in Transformer decoders, known as recency bias. We address this discrepancy by analyzing the interaction between causal self-attention and other architectural components. We show that stacked causal self-attention layers combined with LayerNorm induce recency bias. Furthermore, we examine the effects of residual connections and the distribution of input token embeddings on this bias. Our results provide new theoretical insights into how positional information interacts with architectural components and suggest directions for improving positional encoding strategies.

LayerNorm Induces Recency Bias in Transformer Decoders

Large language models (LLMs) have raised hopes for automated end-to-end fact-checking, but prior studies report mixed results. As mainstream chatbots increasingly ship with reasoning capabilities and web search tools—and millions of users already rely on them for verification—rigorous evaluation is urgent. We evaluate 15 recent LLMs from OpenAI, Google, Meta, and DeepSeek on more than 6,000 claims fact-checked by PolitiFact, comparing standard models with reasoning- and web-search variants. Standard models perform poorly, reasoning offers minimal benefits, and web search provides only moderate gains, despite fact-checks being available on the web. In contrast, a curated RAG system using PolitiFact summaries improved macro F1 by 233% on average across model variants. These findings suggest that giving models access to curated high-quality context is a promising path for automated fact-checking.

Large Language Models Require Curated Context for Reliable Political Fact-Checking—Even with Reasoning and Web Search

The static knowledge representations of large language models (LLMs) inevitably become outdated or incorrect over time. While model-editing techniques offer a promising solution by modifying a model's factual associations, they often produce unpredictable ripple effects, which are unintended behavioral changes that propagate even to the hidden space. In this work, we introduce CLaRE, a lightweight representation-level technique to identify where these ripple effects may occur. Unlike prior gradient-based methods, CLaRE quantifies entanglement between facts using forward activations from a single intermediate layer, avoiding costly backward passes. To enable systematic study, we prepare and analyse a corpus of 11,427 facts drawn from three existing datasets. Using CLaRE, we compute large-scale entanglement graphs of this corpus for multiple models, capturing how local edits propagate through representational space. These graphs enable stronger preservation sets for model editing, audit trails, efficient red-teaming, and scalable post-edit evaluation. In comparison to baselines, CLaRE achieves an average of 62.2% improvement in Spearman correlation with ripple effects while being 2.74× faster, and using 2.85× less peak GPU memory. Besides, CLaRE requires only a fraction of the storage needed by the baselines to compute and preserve fact representations. Our entanglement graphs and corpus are available at https://github.com/manitbaser/CLaRE.

 CLaRE-ty Amid Chaos: Quantifying Representational Entanglement to Predict Ripple Effects in LLM Editing

Multi-turn interaction remains challenging for online reinforcement learning. Current GRPO-based methods—either at the trajectory level or the step level—still suffer from fundamental challenges in multi-turn settings: they allocate sampling uniformly across tasks regardless of difficulty, propagate misleading learning signals that penalize correct intermediate actions in failed trajectories, and incur high sample-collection costs under long-horizon environments. Step-level variants (e.g., GIGPO) mitigate some interaction-cost constraints by decomposing trajectories, yet they retain GRPO’s sampling imbalance and still struggle with heterogeneous multi-turn tasks. To address these issues, we propose STEP (Success-rate-aware Trajectory-Efficient Policy Optimization), a framework that dynamically allocates sampling based on per-task success rates and performs fine-grained step-level optimization. STEP maintains a smoothed success-rate record to guide adaptive trajectory resampling, allocating more effort to harder tasks. It then computes success-rate-weighted advantages and decomposes trajectories into step-level samples, followed by a step-level GRPO augmentation that strengthens updates on low-success tasks. Experiments on OSWorld and AndroidWorld show that STEP substantially improves sample efficiency and training stability over both trajectory-level and existing step-level GRPO variants, converging faster and generalizing better under the same sampling budget.

STEP: Success-Rate-Aware Trajectory-Efficient Policy Optimization

This work investigates the ability of large language models (LLMs) to generate mathematical equations from scientific texts. Prior work faces challenges in unstructured grounding, multi-equation dependency, and human-aligned evaluation. To address this, we construct a dataset of AI research papers, pairing contextual passages with ground-truth equations and variable descriptions. We develop an explainable equation generation workflow and evaluate it across diverse open- and closed-source LLMs. Our evaluation protocol combines automatic metrics, LLM-based rubrics, and human judgments to assess accuracy, explainability, and human-LLM alignment. Results show that LLMs achieve moderate performance on lexical and syntactic similarity, but struggle with semantic accuracy. LLM-based evaluations show limited alignment with human judgments, highlighting challenges in assessing equation quality. These findings provide insights for improving equation generation models and developing more reliable evaluation methods for scientific creativity. We provide code and data for reproducibility.

SciText2Eq: Assessing LLMs for Explainable Equation Generation for Scientific Creativity

While large language models (LLMs) excel at handling long-context sequences, they require substantial prefill computation and key-value (KV) cache, which can heavily burden computational efficiency and memory usage in both prefill and decoding stages.
Recent works that compress KV caches with prefill acceleration reduce this cost but inadvertently tie the prefill compute reduction to the decoding KV budget. This coupling arises from overlooking the layer-dependent variation of critical context, often leading to accuracy degradation. 
To address this issue, we introduce FastKV, a KV cache compression framework designed to reduce latency in both prefill and decoding by leveraging the stabilization of token importance in later layers.
FastKV performs full-context computation until a Token-Selective Propagation (TSP) layer, which forwards only the most informative tokens to subsequent layers.
From these propagated tokens, FastKV independently selects salient KV entries for caching, thereby decoupling KV budget from the prefill compute reduction based on the TSP decision.
This independent control of the TSP rate and KV retention rate enables flexible optimization of efficiency and accuracy.
Experimental results show that FastKV achieves speedups of up to 1.82$\times$ in prefill and 2.87$\times$ in decoding compared to the full-context baseline, while matching the accuracy of the decoding-only baselines.
Our code is available at https://github.com/dongwonjo/FastKV.

Downloads

Next from ACL 2026

Multi-Persona Thinking for Bias Mitigation in Large Language Models

Similar lecture

Brain-resident memory CD8+ T cells provide durable protection against intracranial malignancy