Singapore

Chain‑of‑thought (CoT) prompting boosts Large Language Models accuracy on multi‑step tasks, yet whether the generated ``thoughts&#39;&#39; reflect the true internal reasoning process is unresolved. We present the first feature‑level causal study of CoT faithfulness. Combining sparse autoencoders with activation patching, we extract monosemantic features from Pythia‑70M and Pythia‑2.8B while they tackle GSM8K math problems under CoT and plain (noCoT) prompting. Swapping a small set of CoT‑reasoning features into a noCoT run raises answer log‑probabilities significantly in the 2.8B model, but has no reliable effect in 70M, revealing a clear contrast for these two scales. CoT also leads to significantly higher activation sparsity and feature interpretability scores in the larger model, signalling more modular internal computation. For example, the model&#39;s confidence in generating correct answers improves from 1.2 to 4.3. We introduce patch‑curves and random‑feature patching baselines, showing that useful CoT information is not only present in the top-K patches but widely distributed. Overall, our results indicate that CoT can induce more interpretable internal structures in high-capacity LLMs, validating its role as a structured prompting method.

AAAI 2026

How Does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencoding

Chain‑of‑thought (CoT) prompting boosts Large Language Models accuracy on multi‑step tasks, yet whether the generated ``thoughts'' reflect the true internal reasoning process is unresolved. We present the first feature‑level causal study of CoT faithfulness. Combining sparse autoencoders with activation patching, we extract monosemantic features from Pythia‑70M and Pythia‑2.8B while they tackle GSM8K math problems under CoT and plain (noCoT) prompting. Swapping a small set of CoT‑reasoning features into a noCoT run raises answer log‑probabilities significantly in the 2.8B model, but has no reliable effect in 70M, revealing a clear contrast for these two scales. CoT also leads to significantly higher activation sparsity and feature interpretability scores in the larger model, signalling more modular internal computation. For example, the model's confidence in generating correct answers improves from 1.2 to 4.3. We introduce patch‑curves and random‑feature patching baselines, showing that useful CoT information is not only present in the top-K patches but widely distributed. Overall, our results indicate that CoT can induce more interpretable internal structures in high-capacity LLMs, validating its role as a structured prompting method.

technical paper

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.<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 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 

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.

Ambition and risk-taking have been heralded as important ways for marginalized communities to get out of cycles of poverty. As a result, educational messaging often encourages individuals to strengthen their personal resolve and develop characteristics such as discipline and grit to succeed in ambitious ends. However, recent work in philosophy and sociology highlights that this messaging often does more harm than good for students in these situations. We study similar questions using a different epistemic approach and in simple theoretical models -- we provide a quantitative model of decision-making between stable and risky choices in the improving multi-armed bandits framework. We use this model to first study how individuals' "strategies" are affected by their level of grittiness and how this affects their accrued rewards. Then, we study the impact of various interventions, such as increasing grit or providing a financial safety net. Our investigation of rational decision making involves two different formal models of rationality, the competitive ratio between the accrued reward and the optimal reward and Bayesian quantification of uncertainty.

A Theoretical Model for Grit in Pursuing Ambitious Ends

Reconstructing controllable Gaussian splats for articulated objects from monocular video is especially challenging due to its inherently insufficient constraints. Existing methods address this by relying on dense masks and manually defined control signals, limiting their real-world applications. In this paper, we propose an annotation-free method, **FreeGaussian**, which mathematically disentangles camera egomotion and articulated movements via flow derivatives. By establishing a connection between 2D flows and 3D Gaussian dynamic flow, our method enables optimization and continuity of dynamic Gaussian motions from flow priors without any control signals. Furthermore, we introduce a 3D spherical vector controlling scheme, which represents the state as a 3D Gaussian trajectory, thereby eliminating the need for complex 1D control signal calculations and simplifying controllable Gaussian modeling. Extensive experiments on articulated objects demonstrate the state-of-the-art visual performance and precise, part-aware controllability of our method.

FreeGaussian: Annotation-free Control of Articulated Objects via 3D Gaussian Splats with Flow Derivatives

Data-driven discovery of governing equations from data remains a fundamental challenge in nonlinear dynamics. Although sparse regression techniques have advanced system identification, they struggle with rational functions and noise sensitivity in complex mechanical systems. The Lagrangian formalism offers a promising alternative, as it typically avoids rational expressions and provides a more concise representation of system dynamics. However, existing Lagrangian identification methods are significantly affected by measurement noise and limited data availability. This paper presents a novel differentiable sparse identification framework that addresses these limitations through three key contributions: (1) the first integration of cubic B-Spline approximation into Lagrangian system identification, enabling accurate representation of complex nonlinearities, (2) a robust equation discovery mechanism that effectively utilizes measurements while incorporating known physical constraints, (3) a recursive derivative computation scheme based on B-spline basis functions, effectively constraining higher-order derivatives and reducing noise sensitivity on second-order dynamical systems. The proposed method demonstrates superior performance and enables more accurate and reliable extraction of physical laws from noisy data, particularly in complex mechanical systems compared to baseline methods.

Differentiable Sparse Identification of Lagrangian Dynamics

Road networks are critical infrastructures underpinning intelligent transportation systems and their related applications. Effective representation learning of road networks remains challenging due to the complex interplay between spatial structures and frequency characteristics in traffic patterns. Existing graph neural networks for modeling road networks predominantly fall into two paradigms: spatial-based methods that capture local topology but tend to over-smooth representations, and spectral-based methods that analyze global frequency components but often overlook localized variations. This spatial-spectral misalignment limits their modeling capacity for road networks exhibiting both coarse global trends and fine-grained local fluctuations. To bridge this gap, we propose HiFiNet, a novel hierarchical frequency-decomposition graph neural network that unifies spatial and spectral modeling. HiFiNet constructs a multi-level hierarchy of virtual nodes to enable localized frequency analysis, and employs a decomposition–updating–reconstruction framework with a topology-aware graph transformer to separately model and fuse low- and high-frequency signals. Theoretically justified and empirically validated on multiple real-world datasets across four downstream tasks, HiFiNet demonstrates superior performance and generalization ability in capturing effective road network representations.

Hierarchical Frequency-Decomposition Graph Neural Networks for Road Network Representation Learning

Training Large Language Models (LLMs) for chain-of-thought reasoning presents a significant challenge: supervised fine-tuning on a single "golden" rationale hurts generalization as it penalizes equally valid alternatives, whereas reinforcement learning with verifiable rewards struggles with credit assignment and prohibitive computational cost. To tackle these limitations, we introduce InTRO (\textbf{I}n-\textbf{T}oken \textbf{R}ationality \textbf{O}ptimization), a new framework that enables both token-level exploration and self-feedback for accurate and concise reasoning. Instead of directly optimizing an intractable objective over all valid reasoning paths, InTRO leverages correction factors---token-wise importance weights estimated by the information discrepancy between the generative policy and its answer-conditioned counterpart, for informative next-token selection. This approach allows the model to perform token-level exploration and receive self-generated feedback within a single forward pass, ultimately encouraging accurate and concise rationales. Across six math-reasoning benchmarks, InTRO consistently outperforms other baselines, raising solution accuracy by up to 20\% relative to the base model. Its chains of thought are also notably more concise, exhibiting reduced verbosity. Beyond this, InTRO enables cross-domain transfer, successfully adapting to out-of-domain reasoning tasks that extend beyond the realm of mathematics, demonstrating robust generalization.

In-Token Rationality Optimization: Towards Accurate and Concise LLM Reasoning via Self-Feedback

Text-guided image editing has advanced rapidly with the rise of diffusion models. While flow-based inversion-free methods offer high efficiency by avoiding latent inversion, they often fail to effectively integrate source information, leading to poor background preservation, spatial inconsistencies, and over-editing due to the lack of effective integration of source information.
In this paper, we present $\textbf{FIA-Edit}$, a novel inversion-free framework that achieves high-fidelity and semantically precise edits through a $\textbf{F}$requency-$\textbf{I}$nteractive $\textbf{A}$ttention. Specifically, we design two key components: (1) a Frequency Representation Interaction (FRI) module that enhances cross-domain alignment by exchanging frequency components between source and target features within self-attention, and (2) a Feature Injection (FIJ) module that explicitly incorporates source-side queries, keys, values, and text embeddings into the target branch's cross-attention to preserve structure and semantics. 
Comprehensive and extensive experiments demonstrate that FIA-Edit supports high-fidelity editing at low computational cost ($\sim$6s per $512\times 512$ image on an RTX 4090) and consistently outperforms existing methods across diverse tasks in visual quality, background fidelity, and controllability.
Furthermore, we are the first to extend text-guided image editing to clinical applications. By synthesizing anatomically coherent hemorrhage variations in surgical images, FIA-Edit opens new opportunities for medical data augmentation and delivers significant gains in downstream bleeding classification.

FIA-Edit: Frequency-Interactive Attention for Efficient and High-Fidelity Inversion-Free Text-Guided Image Editing

Understanding true influence in social media requires distinguishing correlation from causation—particularly when analyzing misinformation spread. While existing approaches focus on exposure metrics and network structures, they often fail to capture the causal mechanisms by which external temporal signals trigger engagement. We introduce CITRUS (Causal Influence through Treatment-Response Understanding in Social media), a novel joint treatment-outcome framework that leverages existing sequential models to understand how external signals—search trends, news coverage, influencer activity—trigger misinformation engagement. Through experiments on real-world misinformation and disinformation datasets, CITRUS outperforms existing benchmarks by $15$-$22$\% in predicting engagement across diverse counterfactual scenarios, including exposure adjustment, temporal alignment shifts, and varied intervention durations. 
Case studies on $492$ social media users demonstrate that our causal effect measure aligns strongly with expert-based empirical influence assessments, validating CITRUS as a robust framework for understanding information spread dynamics.
CITRUS also reveals that low-baseline misinformation can scale 6-fold under external promotion, showing super-linear growth, and unmasks hidden amplifiers—accounts with modest followings that double engagement rates, outperforming supposed "influencers" with $100$x more followers.

Estimating Online Influence Needs Causal Modeling! Counterfactual Analysis of Misinformation Engagement on Social Media

Offline Reinforcement Learning (RL) enables policy improvement from fixed datasets without online interactions, making it highly suitable for real-world applications lacking efficient simulators.
Despite its success in the single-agent setting, offline multi-agent RL remains a challenge, especially in competitive games.
Firstly, unaware of the game structure, it is impossible to interact with the opponents and conduct a major learning paradigm, self-play, for competitive games. Secondly, real-world datasets cannot cover all the state and action space in the game, resulting in barriers to identifying Nash equilibrium (NE).
To address these issues, this paper introduces Off-FSP, the first practical model-free offline RL algorithm for competitive games.
We start by simulating interactions with various opponents by adjusting the weights of the fixed dataset with importance sampling.
This technique allows us to learn the best responses to different opponents and employ the Offline Self-Play learning framework. 
To overcome the challenge of partial coverage, we combine the single-agent offline RL method with Fictitious Self-Play (FSP) to approximate NE by constraining the approximate best responses away from out-of-distribution actions.
Experiments on matrix games, extensive-form poker, and board games demonstrate that Off-FSP achieves significantly lower exploitability than state-of-the-art baselines. Finally, we validate Off-FSP on a real-world human-robot competitive task, demonstrating its potential for solving complex, hard-to-simulate real-world problems.

Offline Fictitious Self-Play for Competitive Games

Out-of-Town (OOT) recommendation aims to provide personalized suggestions for users in unfamiliar cities. However, OOT recommendation faces two fundamental challenges: the difficulty of reasoning across modalities, as preference signals in disparate formats such as images and text are hard to compare; and the preference deviation problem, since a user's resident and tourist preferences often diverge, rendering simple preference transfer ineffective. To address these challenges, we propose Distinguishing Resident and Tourist Preferences via Multi-Modal LLM Alignment for Out-of-Town Cross-Domain Recommendation (DiMA), a framework for re-ranking Points of Interest (POIs). To tackle the multimodal challenge, DiMA first leverages Multimodal Large Language Models (MLLMs) and Large Language Models (LLMs) to transform heterogeneous POI data into unified semantic tags, enabling both cross-modal reasoning and efficient downstream processing. To address preference deviation, a ``teacher'' LLM executes a custom Chain-of-Thought (CoT) process to disentangle resident and tourist preferences from multi-city histories for re-ranking. Finally, a lightweight student model learns this CoT reasoning via Supervised Fine-Tuning (SFT) and is then refined with Direct Preference Optimization (DPO) to align with true user choices, with the potential to surpass the teacher. Extensive experiments on a real-world dataset demonstrate that DiMA significantly enhances the performance of baseline models in the OOT recommendation re-ranking task.

DiMA: Distinguishing Resident and Tourist Preferences via Multi-Modal LLM Alignment for Out-of-Town Cross-Domain Recommendation

Medical vision–language pretraining typically relies on static image–text pairs, overlooking temporal cues vital for understanding clinical progression. This limits model sensitivity to evolving semantics and reduces their effectiveness in real-world clinical reasoning. To address this challenge, we propose TAMM—a temporal alignment framework that leverages weak but semantically rich supervision from large language models (LLMs). Given temporally adjacent clinical reports, LLMs automatically generate (i) coarse-grained trend labels (e.g., improving or worsening), and (ii) fine-grained rationales explaining the supporting clinical evidence. These complementary signals inject temporal semantics without requiring manual annotation, and guide vision–language representation learning to capture trend-sensitive cross-modal alignment and rationale-grounded coherence. Experiments on multiple medical benchmarks demonstrate that TAMM improves retrieval and classification performance while yielding more interpretable, temporally consistent embeddings. Our results highlight the potential of leveraging LLM-derived supervision to equip vision–language models with temporal awareness critical for clinical applications.

Downloads

Next from AAAI 2026

A Theoretical Model for Grit in Pursuing Ambitious Ends

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

A Theoretical Model for Grit in Pursuing Ambitious Ends

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads