Getting into ML research these days can be exciting but also overwhelming. In this talk, I will start by sharing my own journey as an NLP researcher in a decade full of "did my problem just get solved?" crises. I will then share my perspective on choosing meaningful research problems and general tips for grad students navigating ML research.

Recent advances in diffusion models have enabled high-quality generation and manipulation of images guided by texts, as well as concept learning from images. However, naive applications of existing methods to editing tasks that require fine-grained control, e.g., face editing, often lead to suboptimal solutions with identity information and high-frequency details lost during the editing process, or irrelevant image regions altered due to entangled concepts. In this work, we propose S²Edit, a novel method based on a pre-trained text-to-image diffusion model that enables personalized editing with precise semantic and spatial control. We first fine-tune our model to embed the identity information into a learnable text token. During fine-tuning, we disentangle the learned identity token from attributes to be edited by enforcing an orthogonality constraint in the textual feature space. To ensure that the identity token only affects regions of interest, we apply object masks to guide the cross-attention maps. At inference time, our method performs localized editing while faithfully preserving the original identity with semantically disentangled and spatially focused identity token learned. Extensive experiments demonstrate the superiority of S²Edit over state-of-the-art methods both quantitatively and qualitatively. Additionally, we showcase several compositional image editing applications of S²Edit such as makeup transfer.

Modeling user behavior is critical across many industries where understanding preferences, intent, or decisions informs personalization, targeting, and strategic outcomes. Surveys have long served as a classical mechanism for collecting such behavioral data due to their interpretability, structure, and ease of deployment. However, surveys alone are inherently limited by user fatigue, incomplete responses, and practical constraints on their length making them insufficient for capturing user behavior. In this work, we present LANTERN (Late-Attentive Network for Enriched Response Modeling), a modular architecture for modeling user behavior by fusing adaptive survey responses with supplemental contextual signals. We demonstrate the architectural value of maintaining survey primacy through selective gating, residual connections and late fusion via cross-attention, treating survey data as the primary signal while incorporating external modalities only when relevant. LANTERN outperforms strong survey-only baselines in multi-label prediction of survey responses. We further investigate threshold sensitivity and the benefits of selective modality reliance through ablation and rare/frequent attribute analysis. LANTERN is currently deployed in production, where its modularity supports scalable integration of new encoders and evolving datasets. This work provides a practical and extensible blueprint for behavior modeling in survey-centric applications.

Reinforcement Learning (RL) has become a new engine for recent AI breakthroughs, from RLHF and Reasoning Training to Agentic Models. In this talk, we will cover the basics of RL training for LLMs and share our experiences on developing our large-scale RL system AReaL, for LLMs. AReaL is a fully asynchronous training system for LLMs and agents. The AReaL project releases a collection of reproducible recipes for building SOTA LLM applications, including all the training data, model and scripts. We hope AReaL can help researchers build their own LLM models and agents easily. Our projects can be found at https://github.com/inclusionAI/AReaL.

Join me as I take you through the twists, turns, and transformative moments of my PhD journey—beyond the papers and deadlines. Think of it as a preview for those considering a PhD, a survival guide (and a cautionary tale!) for those in their early years, and nostalgia for those who've made it through and lived to tell the tale!

As the development of artificial general intelligence (AGI) accelerates, questions of societal impact are becoming increasingly urgent. This talk explores how AGI could interface with human governance structures, democratic institutions, and pluralistic values. I'll outline key risks posed by AGI concentration—such as misalignment, surveillance, and political capture—and explore frameworks for ensuring AGI benefits are distributed justly. Drawing on lessons from political philosophy and social choice theory, I'll propose a vision of AGI as both a tool of public infrastructure and a mechanism for holding institutions accountable. The talk calls for machine learning researchers to engage with political legitimacy and the design of global participatory systems.

Pretrained molecular encoders have become indispensable in computational chemistry for tasks such as property prediction and molecular generation. However, the standard practice of relying solely on final-layer embeddings for downstream tasks may discard valuable information. In this work, we challenge this convention by conducting a comprehensive layer-wise analysis of five diverse molecular encoders across 22 ADMET property prediction tasks. Our results demonstrate that embeddings from intermediate layers consistently outperform final-layer representations: Using fixed embeddings from the optimal intermediate layers improved downstream performance by an average of 5.4%, reaching gains up to 28.6%; finetuning up to these intermediate layers yielded even greater average improvements of 8.5%, with performance increases as high as 40.8%, achieving new state-of-the-art results on several benchmarks. These findings highlight the importance of exploring the full representational depth of molecular encoders to achieve substantial performance improvements and computational efficiency.

In this paper, we propose ScoreMix, a novel yet simple data augmentation strategy leveraging the score compositional properties of diffusion models to enhance discriminator performance, particularly under scenarios with limited labeled data. By convexly mixing the scores from different class-conditioned trajectories during diffusion sampling, we generate challenging synthetic samples that significantly improve discriminative capabilities in all studied benchmarks. We systematically investigate class-selection strategies for mixing and discover that greater performance gains arise when combining classes distant in the discriminator's embedding space, rather than close in the generator's condition space. Moreover, we empirically show that, under standard metrics, the correlation between the generator's learned condition space and the discriminator's embedding space is minimal. Our approach achieves notable performance improvements without extensive parameter searches, demonstrating practical advantages for training discriminative models while effectively mitigating problems regarding collections of large datasets. The source code and the synthetic dataset will be publicly available when published.