Singapore

Alignment plays a crucial role in Large Language Models (LLMs) in aligning with human preferences on a specific task/domain. Traditional alignment methods suffer from catastrophic forgetting, where models lose previously learned values when adapting to new preferences or domains. We introduce LifeAlign, a novel framework for lifelong alignment that enables LLMs to maintain consistent human preference alignment across sequential learning tasks without forgetting previously learned values. Our approach consists of two key innovations. First, we propose a focalized preference optimization strategy that aligns LLMs with new preferences while preventing the erosion of alignment acquired from previous tasks. Second, we develop a short-to-long memory consolidation mechanism that merges denoised short-term preference representations into stable long-term memory using intrinsic dimensionality reduction, enabling efficient storage and retrieval of alignment patterns across diverse domains. We evaluate LifeAlign across multiple sequential alignment tasks spanning different domains and preference types. Experimental results demonstrate that our method achieves superior performance in maintaining both preference alignment quality and knowledge retention compared to existing lifelong learning approaches.

AAAI 2026

LifeAlign: Lifelong Alignment for Large Language Models with Memory-Augmented Focalized Preference Optimization

life-long and continual learning

alignment

large language models

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community." 

*Message from the General Chair, Dirk Hovy*

EMNLP 2025

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

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.

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 

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) are transforming the field of
natural language processing, yet their development remains
concentrated on a handful of high-resource languages,
raising fundamental questions of inclusivity, trust, and
global accessibility. My research addresses these
challenges by advancing multilingual and trustworthy AI. On
the multilingual front, I have analyzed how LLMs internally
process diverse languages, introduced benchmarks such as
M3Exam and SeaBench to reveal performance gaps, and led
large-scale open-source initiatives including SeaLLMs and
Babel that extend strong model support to underrepresented
languages worldwide. Complementing inclusivity, my work
also uncovers vulnerabilities in LLMs (e.g., multilingual
jailbreaks) and introduces neuron-level interpretability
and automated evaluation frameworks (e.g., Auto-Arena) for
trustworthy deployment. Looking ahead, I aim to build AI
systems that are linguistically inclusive, culturally
aware, and inherently safe, bridging foundational advances
with real-world applications in diverse global contexts.

Towards Inclusive AI: Advancing Multilingual Large Language Models

Slum segmentation from satellite imagery holds significant promise in generating consistent global estimates of urban poverty. However, the morphological heterogeneity of informal settlements presents a major challenge, limiting the generalization of models trained on specific regions to unseen locations. To address this, we introduce a large-scale high-resolution dataset and propose GRAM (Generalized Region-Aware Mixture-of-Experts), a two-phase test-time adaptation framework that enables robust slum segmentation without labeled data from target regions. We compile a million-scale dataset of preprocessed satellite imagery from 12 cities across four continents for source training. Using this data set, GRAM employs a mixture-of-experts architecture to capture region-specific slum characteristics while learning universal features through a shared backbone. During adaptation, prediction consistency across experts filters unreliable pseudo-labels, allowing the model to generalize effectively to previously unseen regions. When tested on three African cities, GRAM outperforms state-of-the-art baselines in low-resource settings, offering a scalable and label-efficient solution for global slum mapping and data-driven urban planning.

Generalizable Slum Detection from Satellite Imagery with Mixture-of-Experts

This paper presents an eXplainable AI (XAI)-based class-
room game “Breakable Machine” for teaching critical, trans-
formative AI literacy through adversarial play and interroga-
tion of AI systems. Designed for learners aged 10–15, the
game invites students to spoof an image classifier by manip-
ulating their appearance or environment in order to trigger
high-confidence misclassifications. Rather than focusing on
building AI models, this activity centers on breaking them—
exposing their brittleness, bias, and vulnerability through
hands-on, embodied experimentation. The game includes an
XAI view to help students visualize feature saliency, reveal-
ing how models attend to specific visual cues. A shared class-
room leaderboard fosters collaborative inquiry and compar-
ison of strategies, turning the classroom into a site for col-
lective sensemaking. This approach repositions AI education
by treating model failure and misclassification not as prob-
lems to be debugged, but as pedagogically rich opportuni-
ties to interrogate AI as a sociotechnical system. In doing so,
the game supports students in developing data agency, ethical
awareness, and a critical stance toward AI systems increas-
ingly embedded in everyday life. The game and its source
code are freely available.

Breakable Machine: A K–12 Classroom Game for Transformative AI Literacy Through Spoofing and eXplainable AI (XAI)

Commercial-grade poster design demands the seamless integration of aesthetic appeal with precise, informative content delivery. Current automated poster generation systems face significant limitations, including incomplete design workflows, poor text rendering accuracy, and insufficient flexibility for commercial applications. To address these challenges, we propose PosterVerse, a full-workflow, commercial-grade poster generation method that seamlessly automates the entire design process while delivering high-density and scalable text rendering. PosterVerse replicates professional design through three key stages: (1) blueprint creation using fine-tuned LLMs to extract key design elements from user requirements, (2) graphical background generation via customized diffusion models to create visually appealing imagery, and (3) unified layout-text rendering with an MLLM-powered HTML engine to guarantee high text accuracy and flexible customization. In addition, we introduce PosterDNA, a commercial-grade, HTML-based dataset tailored for training and validating poster design models. To the best of our knowledge, PosterDNA is the first Chinese poster generation dataset to introduce HTML typography files, enabling scalable text rendering and fundamentally solving the challenges of rendering small and high-density text. Experimental results demonstrate that PosterVerse consistently produces commercial-grade posters with appealing visuals, accurate text alignment, and customizable layouts, making it a promising solution for automating commercial poster design. Dataset and code will be publicly available.

PosterVerse: A Full-Workflow Framework for Commercial-Grade Poster Generation with HTML-Based Scalable Typography

We study streaming data with categorical features where the universe of distinct categorical items is not known in advance and can even grow unboundedly over time.
Feature hashing is commonly used as a pre-processing step to map these categorical values into a feature space of fixed size before learning their embeddings \citep{coleman2023unified, desai2022random}. While these methods have been developed and evaluated for offline or batch settings, in this paper we consider online settings. 
We show that deterministic embeddings suffer from forgetting in online learning, leading to performance deterioration. 
To mitigate this issue, we propose a \textit{probabilistic hash embedding} (PHE) model that treats hash embeddings as stochastic and applies Bayesian online learning to learn incrementally from data. 
Based on the structure of PHE, we derive a scalable inference algorithm to learn model parameters and infer/update the posteriors of hash embeddings and other latent variables. 
Our algorithm (i) can handle an evolving vocabulary of categorical items, (ii) is adaptive to new items without forgetting old items, (iii) is implementable with a bounded set of parameters that does not grow with the number of distinct observed values on the stream, and (iv) is efficiently implementable both in the offline and the online streaming setting. 
Experiments in classification, sequence modeling, and recommendation systems in online learning setups demonstrate the superior performance of PHE while maintaining high memory efficiency (consumes as low as 2~4% memory of a one-hot embedding table).

Probabilistic Hash Embeddings for Online Learning of Categorical Features

The long-term progression of neurodegenerative diseases is commonly conceptualized as a spatiotemporal diffusion process that consists of a graph diffusion process across the structural brain connectome and a localized reaction process within brain regions. However, modeling this progression remains challenging due to 1) the scarcity of longitudinal data obtained through irregular and infrequent subject visits and 2) the complex interplay of pathological mechanisms across brain regions and disease stages, where traditional models assume fixed mechanisms throughout disease progression. To address these limitations, we propose a novel stage-aware Mixture of Experts (MoE) framework that explicitly models how different contributing mechanisms dominate at different disease stages through time-dependent expert weighting. This architecture is a key innovation designed to maximize the utility of small datasets and provide interpretable insights into disease etiology. Data-wise, we utilize an iterative dual optimization method to properly estimate the temporal position of individual observations, constructing a cohort-level progression trajectory from irregular snapshots. Model-wise, we enhance the spatial component with an inhomogeneous graph neural diffusion model (IGND) that allows diffusivity to vary based on node states and time, providing more flexible representations of brain networks. We also introduce a localized neural reaction module to capture complex dynamics beyond standard processes.The resulting IGND-MoE model dynamically integrates these components across temporal states, offering a principled way to understand how stage-specific pathological mechanisms contribute to progression. When used to model tau pathology propagation in human brains, IGND-MoE outperforms purely pathophysiological and purely neural baselines in long-term prediction accuracy. Moreover, its stage-wise weights yield novel clinical insights that align with literature, suggesting that graph-related processes are more influential at early stages, while other unknown physical processes become dominant later on. Our findings highlight the necessity of designing hybrid and expert-constrained models that account for the evolving nature of neurodegenerative processes.

A Stage-Aware Mixture of Experts Framework for Neurodegenerative Disease Progression Modelling

Recent self-supervised image segmentation models have achieved promising performance on semantic segmentation and class-agnostic instance segmentation. 
However, their pretraining schedule is multi-stage, requiring a time-consuming pseudo-masks generation process between each training epoch. 
This time-consuming offline process not only makes it difficult to scale with training dataset size, but also leads to sub-optimal solutions due to its discontinuous optimization routine. 
To solve these, we first present a novel pseudo-mask algorithm, Fast Universal Agglomerative Pooling (UniAP). Each layer of UniAP can identify groups of similar nodes in parallel, allowing to generate both semantic-level and instance-level and multi-granular pseudo-masks within ens of milliseconds for one image. 
Based on the fast UniAP, we propose the Scalable Self-Supervised Universal Segmentation (S2-UniSeg), which employs a student and a momentum teacher for continuous pretraining. 
A novel segmentation-oriented pretext task, Query-wise Self-Distillation (QuerySD), is proposed to pretrain S2-UniSeg to learn the local-to-global correspondences. 
Under the same setting, S2-UniSeg outperforms the SOTA UnSAM model, achieving notable improvements of AP+6.9 on COCO, AR+11.1 on UVO, PixelAcc+4.5 on COCOStuff-27, RQ+8.0 on Cityscapes. After scaling up to a larger 2M-image subset of SA-1B, S2-UniSeg further achieves performance gains on all four benchmarks. 
Our code and pretrained models shall be released upon the acceptance of this work.

S2-UniSeg: Fast Universal Agglomerative Pooling for Scalable Segment Anything Without Supervision

State-of-the-art fact-checking systems combat misinformation at scale by employing autonomous LLM-based agents to decompose complex claims into smaller sub-claims, verify each sub-claim individually, and aggregate the partial results to produce verdicts with justifications (explanatory rationales for the verdicts).
The security of these systems is crucial, as compromised fact-checkers, which tend to be easily underexplored, can amplify misinformation.
This work introduces Fact2Fiction, the first poisoning attack framework targeting such agentic fact-checking systems.
Fact2Fiction mirrors the decomposition strategy and exploits system-generated justifications to craft tailored malicious evidences that compromise sub-claim verification.
Extensive experiments demonstrate that Fact2Fiction achieves 8.9\%--21.2\% higher attack success rates than state-of-the-art attacks across various poisoning budgets. Fact2Fiction exposes security weaknesses in current fact-checking systems and highlights the need for defensive countermeasures.

Fact2Fiction: Targeted Poisoning Attack to Agentic Fact-checking System

Quantum computing is finding promising applications in optimization, machine learning and physics, leading to the development of various models for representing quantum information classically.
Because these representations are often studied in different contexts (many-body physics, machine learning, formal verification, simulation), little is known about fundamental trade-offs between their succinctness and the runtime of operations to update them.
We therefore analytically investigate widely-used quantum state representations: matrix product states (MPS), several decision diagram (DD) variants, and restricted Boltzmann machines (RBMs).
We map the relative succinctness of these data structures and provide the complexity for relevant query and manipulation operations. Further, we introduce ``rapidity'' for non-canonical representations: representation $A$ is at least as rapid as $B$, if it can polynomially simulate an operation performed on $B$ \emph{taking the size of $B$ as a unit}.
We then provide a framework for deriving rapidity relations for almost all operations in one stroke.
Our results show inter alia 
(1) that most DD variants are redundant with respect to MPS in the strongest asymptotic sense
(MPS is at least as rapid as those DD variants for almost all operations),
(2) that only one DD variant called LIMDD and RBM have succinctness incomparable to MPS, and
(3) that LIMDD and RBM seem to achieve this by sacrificing tractability of counting queries through a metatheorem on conditional hardness of these queries.

A Knowledge Compilation Map for Quantum Information

We train Transformer-based language models on ten foundational algorithmic tasks and observe pronounced phase transitions in their loss curves that deviate from established power-law scaling trends. Over large ranges of compute, the validation loss barely improves, then abruptly decreases. Probing the models’ internal representations reveals that quiet features are learned prior to any decrease in task loss. These quiet features represent intermediate algorithmic computations that do not by themselves improve the output loss. Ablation experiments demonstrate that individual quiet features are causally necessary for task performance. Our results demonstrate that substantial representational progress can remain hidden beneath an apparently flat loss curve, challenging the prevailing use of cross‑entropy as a proxy for learning and motivating richer diagnostics for monitoring model training.

Downloads

Next from AAAI 2026

Towards Inclusive AI: Advancing Multilingual Large Language Models

Similar lecture

Uncovering Factor-Level Preference to Improve Human-Model Alignment

.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

Towards Inclusive AI: Advancing Multilingual Large Language Models

Similar lecture

Uncovering Factor-Level Preference to Improve Human-Model Alignment

Downloads