Chao-Chun Hsu

Chao-Chun Hsu, Erin Bransom, Jenna Sparks et al. · allen-ai, uw

Literature review requires researchers to synthesize a large amount of information and is increasingly challenging as the scientific literature expands. In this work, we investigate the potential of LLMs for producing hierarchical organizations of scientific studies to assist researchers with literature review. We define hierarchical organizations as tree structures where nodes refer to topical categories and every node is linked to the studies assigned to that category. Our naive LLM-based pipeline for hierarchy generation from a set of studies produces promising yet imperfect hierarchies, motivating us to collect CHIME, an expert-curated dataset for this task focused on biomedicine. Given the challenging and time-consuming nature of building hierarchies from scratch, we use a human-in-the-loop process in which experts correct errors (both links between categories and study assignment) in LLM-generated hierarchies. CHIME contains 2,174 LLM-generated hierarchies covering 472 topics, and expert-corrected hierarchies for a subset of 100 topics. Expert corrections allow us to quantify LLM performance, and we find that while they are quite good at generating and organizing categories, their assignment of studies to categories could be improved. We attempt to train a corrector model with human feedback which improves study assignment by 12.6 F1 points. We release our dataset and models to encourage research on developing better assistive tools for literature review.

Manan Suri, Xiangci Li, Mehdi Shojaie et al. · amazon-science

Current AI-powered code assistance tools often struggle with poorly-defined problem statements that lack sufficient task context and requirements specification. Recent analysis of software engineering agents reveals that failures on such underspecified requests are highly correlated with longer trajectories involving either over-exploration or repeated attempts at applying the same fix without proper evolution or testing, leading to suboptimal outcomes across software development tasks. We introduce CodeScout, a contextual query refinement approach that systematically converts underspecified user requests into comprehensive, actionable problem statements through lightweight pre-exploration of the target codebase. Our key innovation is demonstrating that structured analysis before task execution can supplement existing agentic capabilities without requiring any modifications to their underlying scaffolds. CodeScout performs targeted context scoping, conducts multi-perspective analysis examining potential fixes and exploration opportunities, then synthesizes these insights into enhanced problem statements with reproduction steps, expected behaviors, and targeted exploration hints. This pre-exploration directly addresses the identified failure patterns by reducing non-converging agent trajectories while clarifying user intent in natural language space. We evaluate CodeScout using state-of-the-art agentic scaffolds and language models on SWEBench-Verified, demonstrating a 20\% improvement in resolution rates with up to 27 additional issues resolved compared to the default baseline method. Our results suggest that systematic query refinement through contextual analysis represents a promising direction for enhancing AI code assistance capabilities.

Chao-Chun Hsu, Zi-Yuan Chen, Chi-Yang Hsu et al.

Stories are diverse and highly personalized, resulting in a large possible output space for story generation. Existing end-to-end approaches produce monotonous stories because they are limited to the vocabulary and knowledge in a single training dataset. This paper introduces KG-Story, a three-stage framework that allows the story generation model to take advantage of external Knowledge Graphs to produce interesting stories. KG-Story distills a set of representative words from the input prompts, enriches the word set by using external knowledge graphs, and finally generates stories based on the enriched word set. This distill-enrich-generate framework allows the use of external resources not only for the enrichment phase, but also for the distillation and generation phases. In this paper, we show the superiority of KG-Story for visual storytelling, where the input prompt is a sequence of five photos and the output is a short story. Per the human ranking evaluation, stories generated by KG-Story are on average ranked better than that of the state-of-the-art systems. Our code and output stories are available at https://github.com/zychen423/KE-VIST.

Chao-Chun Hsu, Ziad Obermeyer, Chenhao Tan

Physicians write notes about patients. In doing so, they reveal much about themselves. Using data from 129,228 emergency room visits, we train a model to identify notes written by fatigued physicians -- those who worked 5 or more of the prior 7 days. In a hold-out set, the model accurately identifies notes written by these high-workload physicians, and also flags notes written in other high-fatigue settings: on overnight shifts, and after high patient volumes. Model predictions also correlate with worse decision-making on at least one important metric: yield of testing for heart attack is 18% lower with each standard deviation increase in model-predicted fatigue. Finally, the model indicates that notes written about Black and Hispanic patients have 12% and 21% higher predicted fatigue than Whites -- larger than overnight vs. daytime differences. These results have an important implication for large language models (LLMs). Our model indicates that fatigued doctors write more predictable notes. Perhaps unsurprisingly, because word prediction is the core of how LLMs work, we find that LLM-written notes have 17% higher predicted fatigue than real physicians' notes. This indicates that LLMs may introduce distortions in generated text that are not yet fully understood.

Chao-Chun Hsu, Chenhao Tan

Relevance in summarization is typically defined based on textual information alone, without incorporating insights about a particular decision. As a result, to support risk analysis of pancreatic cancer, summaries of medical notes may include irrelevant information such as a knee injury. We propose a novel problem, decision-focused summarization, where the goal is to summarize relevant information for a decision. We leverage a predictive model that makes the decision based on the full text to provide valuable insights on how a decision can be inferred from text. To build a summary, we then select representative sentences that lead to similar model decisions as using the full text while accounting for textual non-redundancy. To evaluate our method (DecSum), we build a testbed where the task is to summarize the first ten reviews of a restaurant in support of predicting its future rating on Yelp. DecSum substantially outperforms text-only summarization methods and model-based explanation methods in decision faithfulness and representativeness. We further demonstrate that DecSum is the only method that enables humans to outperform random chance in predicting which restaurant will be better rated in the future.

Chao-Chun Hsu, Eric Lind, Luca Soldaini et al.

Recent advancements in transformer-based models have greatly improved the ability of Question Answering (QA) systems to provide correct answers; in particular, answer sentence selection (AS2) models, core components of retrieval-based systems, have achieved impressive results. While generally effective, these models fail to provide a satisfying answer when all retrieved candidates are of poor quality, even if they contain correct information. In AS2, models are trained to select the best answer sentence among a set of candidates retrieved for a given question. In this work, we propose to generate answers from a set of AS2 top candidates. Rather than selecting the best candidate, we train a sequence to sequence transformer model to generate an answer from a candidate set. Our tests on three English AS2 datasets show improvement up to 32 absolute points in accuracy over the state of the art.

Jui-Te Huang, Chen-Lung Lu, Po-Kai Chang et al.

Deep reinforcement learning (RL), where the agent learns from mistakes, has been successfully applied to a variety of tasks. With the aim of learning collision-free policies for unmanned vehicles, deep RL has been used for training with various types of data, such as colored images, depth images, and LiDAR point clouds, without the use of classic map--localize--plan approaches. However, existing methods are limited by their reliance on cameras and LiDAR devices, which have degraded sensing under adverse environmental conditions (e.g., smoky environments). In response, we propose the use of single-chip millimeter-wave (mmWave) radar, which is lightweight and inexpensive, for learning-based autonomous navigation. However, because mmWave radar signals are often noisy and sparse, we propose a cross-modal contrastive learning for representation (CM-CLR) method that maximizes the agreement between mmWave radar data and LiDAR data in the training stage. We evaluated our method in real-world robot compared with 1) a method with two separate networks using cross-modal generative reconstruction and an RL policy and 2) a baseline RL policy without cross-modal representation. Our proposed end-to-end deep RL policy with contrastive learning successfully navigated the robot through smoke-filled maze environments and achieved better performance compared with generative reconstruction methods, in which noisy artifact walls or obstacles were produced. All pretrained models and hardware settings are open access for reproducing this study and can be obtained at https://arg-nctu.github.io/projects/deeprl-mmWave.html

Chao-Chun Hsu, Shantanu Karnwal, Sendhil Mullainathan et al.

Machine learning models depend on the quality of input data. As electronic health records are widely adopted, the amount of data in health care is growing, along with complaints about the quality of medical notes. We use two prediction tasks, readmission prediction and in-hospital mortality prediction, to characterize the value of information in medical notes. We show that as a whole, medical notes only provide additional predictive power over structured information in readmission prediction. We further propose a probing framework to select parts of notes that enable more accurate predictions than using all notes, despite that the selected information leads to a distribution shift from the training data ("all notes"). Finally, we demonstrate that models trained on the selected valuable information achieve even better predictive performance, with only 6.8% of all the tokens for readmission prediction.

Yun-Wei Chu, Kuan-Yen Lin, Chao-Chun Hsu et al.

Understanding dynamic scenes and dialogue contexts in order to converse with users has been challenging for multimodal dialogue systems. The 8-th Dialog System Technology Challenge (DSTC8) proposed an Audio Visual Scene-Aware Dialog (AVSD) task, which contains multiple modalities including audio, vision, and language, to evaluate how dialogue systems understand different modalities and response to users. In this paper, we proposed a multi-step joint-modality attention network (JMAN) based on recurrent neural network (RNN) to reason on videos. Our model performs a multi-step attention mechanism and jointly considers both visual and textual representations in each reasoning process to better integrate information from the two different modalities. Compared to the baseline released by AVSD organizers, our model achieves a relative 12.1% and 22.4% improvement over the baseline on ROUGE-L score and CIDEr score.

Yi-Wei Huang, Tzu-Kuan Chuang, Ni-Ching Lin et al.

Robotic software and hardware systems of autonomous surface vehicles have been developed in transportation, military, and ocean researches for decades. Previous efforts in RobotX Challenges 2014 and 2016 facilitates the developments for important tasks such as obstacle avoidance and docking. Team NCTU is motivated by the AI Driving Olympics (AI-DO) developed by the Duckietown community, and adopts the principles to RobotX challenge. With the containerization (Docker) and uniformed AI agent (with observations and actions), we could better 1) integrate solutions developed in different middlewares (ROS and MOOS), 2) develop essential functionalities of from simulation (Gazebo) to real robots (either miniaturized or full-sized WAM-V), and 3) compare different approaches either from classic model-based or learning-based. Finally, we setup an outdoor on-surface platform with localization services for evaluation. Some of the preliminary results will be presented for the Team NCTU participations of the RobotX competition in Hawaii in 2018.

Kuan-Yen Lin, Chao-Chun Hsu, Yun-Nung Chen et al.

With increasing information from social media, there are more and more videos available. Therefore, the ability to reason on a video is important and deserves to be discussed. TheDialog System Technology Challenge (DSTC7) (Yoshino et al. 2018) proposed an Audio Visual Scene-aware Dialog (AVSD) task, which contains five modalities including video, dialogue history, summary, and caption, as a scene-aware environment. In this paper, we propose the entropy-enhanced dynamic memory network (DMN) to effectively model video modality. The attention-based GRU in the proposed model can improve the model's ability to comprehend and memorize sequential information. The entropy mechanism can control the attention distribution higher, so each to-be-answered question can focus more specifically on a small set of video segments. After the entropy-enhanced DMN secures the video context, we apply an attention model that in-corporates summary and caption to generate an accurate answer given the question about the video. In the official evaluation, our system can achieve improved performance against the released baseline model for both subjective and objective evaluation metrics.

Chao-Chun Hsu, Yu-Hua Chen, Zi-Yuan Chen et al.

In this paper, we introduce Dixit, an interactive visual storytelling system that the user interacts with iteratively to compose a short story for a photo sequence. The user initiates the process by uploading a sequence of photos. Dixit first extracts text terms from each photo which describe the objects (e.g., boy, bike) or actions (e.g., sleep) in the photo, and then allows the user to add new terms or remove existing terms. Dixit then generates a short story based on these terms. Behind the scenes, Dixit uses an LSTM-based model trained on image caption data and FrameNet to distill terms from each image and utilizes a transformer decoder to compose a context-coherent story. Users change images or terms iteratively with Dixit to create the most ideal story. Dixit also allows users to manually edit and rate stories. The proposed procedure opens up possibilities for interpretable and controllable visual storytelling, allowing users to understand the story formation rationale and to intervene in the generation process.

Chao-Chun Hsu, Szu-Min Chen, Ming-Hsun Hsieh et al.

Visual storytelling includes two important parts: coherence between the story and images as well as the story structure. For image to text neural network models, similar images in the sequence would provide close information for story generator to obtain almost identical sentence. However, repeatedly narrating same objects or events will undermine a good story structure. In this paper, we proposed an inter-sentence diverse beam search to generate a more expressive story. Comparing to some recent models of visual storytelling task, which generate story without considering the generated sentence of the previous picture, our proposed method can avoid generating identical sentence even given a sequence of similar pictures.

Sheng-Yeh Chen, Chao-Chun Hsu, Chuan-Chun Kuo et al.

Feeling emotion is a critical characteristic to distinguish people from machines. Among all the multi-modal resources for emotion detection, textual datasets are those containing the least additional information in addition to semantics, and hence are adopted widely for testing the developed systems. However, most of the textual emotional datasets consist of emotion labels of only individual words, sentences or documents, which makes it challenging to discuss the contextual flow of emotions. In this paper, we introduce EmotionLines, the first dataset with emotions labeling on all utterances in each dialogue only based on their textual content. Dialogues in EmotionLines are collected from Friends TV scripts and private Facebook messenger dialogues. Then one of seven emotions, six Ekman's basic emotions plus the neutral emotion, is labeled on each utterance by 5 Amazon MTurkers. A total of 29,245 utterances from 2,000 dialogues are labeled in EmotionLines. We also provide several strong baselines for emotion detection models on EmotionLines in this paper.