Ya Gao

Enabling artificial intelligence systems, particularly large language models, to integrate new knowledge and flexibly apply it during reasoning remains a central challenge. Existing knowledge editing approaches emphasize atomic facts, improving factual recall but often failing to integrate new information into a coherent framework usable across contexts. In this work, we argue that knowledge internalization is fundamentally a reasoning problem rather than a memorization problem. Consequently, a model should be trained in situations where the new information is instrumental to solving a task, combined with pre-existing knowledge, and exercised through multi-step reasoning. Based on this insight, we propose a training strategy based on three principles. First, new knowledge is introduced as a coherent background story that contextualizes novel facts and explains their relation to existing knowledge. Second, models are trained using self-generated multi-hop questions that require multi-step reasoning involving the new information. Third, training is done using knowledge distillation, forcing a student model to internalize the teacher's reasoning behavior without access to the novel information. Experiments show that models trained with this strategy effectively leverage newly acquired knowledge during reasoning and achieve remarkable performance on challenging questions that require combining multiple new facts.

Large Vision-Language Models (LVLMs) often suffer from object hallucination, generating text inconsistent with visual inputs, which can critically undermine their reliability. Existing inference-time interventions to mitigate this issue present a challenging trade-off: while methods that steer internal states or adjust output logits can be effective, they often incur substantial computational overhead, typically requiring extra forward passes. This efficiency bottleneck can limit their practicality for real-world, latency-sensitive deployments. In this work, we aim to address this trade-off with Residual-Update Directed DEcoding Regulation (RUDDER), a low-overhead framework that steers LVLMs towards visually-grounded generation. RUDDER is built on two key innovations: (1) Contextual Activation Residual Direction (CARD) vector, a per-sample visual evidence vector extracted from the residual update of a self-attention layer during a single, standard forward pass. (2) A Bayesian-inspired adaptive gate that performs token-wise injection, applying a corrective signal whose strength is conditioned on the model's deviation from the visual context. Extensive experiments on key hallucination benchmarks, including POPE and CHAIR, indicate that RUDDER achieves performance comparable to state-of-the-art methods while introducing negligible computational latency, validating RUDDER as a pragmatic and effective approach for improving LVLMs' reliability without a significant compromise on efficiency.

Digital phenotyping, which takes advantage of data continuously gathered from smartphones and wearable devices, offers promising avenues for real-time monitoring and mental health analysis. This approach holds promise for improving early detection and personalized care in mood disorders by enabling clinicians to proactively respond to significant changes before symptoms worsen. However, the complexity and heterogeneity of digital phenotyping data pose significant modeling challenges. Recent advances in large language models (LLMs) suggest their potential to generalize across diverse tasks with minimal labeled data, making them a promising alternative for analyzing data from digital phenotyping studies. However, the extent of usability of these methods for digital phenotyping studies is not yet well understood. In this study, we evaluate the potential of LLMs in analyzing digital phenotyping data to predict changes in depression severity among individuals experiencing major depressive episodes. We evaluate several in-context learning and fine-tuning strategies, and find that both few-shot prompted LLMs and fine-tuned models outperform traditional machine learning baselines trained on the same set of input features. Moreover, we compare two fine-tuning approaches (fine-tuning only the embedding layer versus parameter-efficient fine-tuning using QLoRA) and find that fine-tuning only the embedding layer significantly improves performance compared to QLoRA fine-tuning. These results highlight the capability of LLMs to process and integrate heterogeneous behavioral data, promising their application for digital phenotyping and mental health research. While our findings highlight the potential in using LLMs in mental health monitoring, their black-box nature and risk of replicating data biases highlight the need for clinical oversight and validation in real-world practice.

An adverse drug event (ADE) is defined as an adverse reaction resulting from improper drug use, reported in various documents such as biomedical literature, drug reviews, and user posts on social media. The recent advances in natural language processing techniques have facilitated automated ADE detection from documents. However, the contextualized information and relations among text pieces are less explored. This paper investigates contextualized language models and heterogeneous graph representations. It builds a contextualized graph embedding model for adverse drug event detection. We employ different convolutional graph neural networks and pre-trained contextualized embeddings as the building blocks. Experimental results show that our methods can improve the performance by comparing recent ADE detection models, suggesting that a text graph can capture causal relationships and dependency between different entities in a document.