publications

2024

1. Rate, Explain and Cite (REC): Enhanced Explanation and Attribution in Automatic Evaluation by Large Language Models

   Aliyah R. Hsu, James Zhu, Zhichao Wang, Bin Bi, Shubham Mehrotra, Shiva Kumar Pentyala, Katherine Tan, Xiang-Bo Mao, Roshanak Omrani, Sougata Chaudhuri, Regunathan Radhakrishnan, Sitaram Asur, Claire Na Cheng, and Bin Yu

   In Preprint, 2024

   PDF
2. A Generative Framework to Bridge Data-driven Models and Scientific Theories in Language Neuroscience

   Richard Antonello*, Chandan Singh*, Shailee Jain, Aliyah R. Hsu, Jianfeng Gao, Bin Yu, and Alexander Huth

   In Preprint, 2024

   PDF
3. Efficient Automated Circuit Discovery in Transformers using Contextual Decomposition

   Aliyah R. Hsu, Georgia Zhou, Yeshwanth Cherapanamjeri, Yaxuan Huang, Anobel Odisho, Peter R. Carroll, and Bin Yu

   In Preprint, 2024

   PDF
4. Diagnosing Transformers: Illuminating Feature Spaces for Clinical Decision-Making

   Aliyah R. Hsu, Yeshwanth Cherapanamjeri, Briton Park, Tristan Naumann, Anobel Odisho, and Bin Yu

   In The Twelfth International Conference on Learning Representations (ICLR), 2024

   PDF

2023

1. Explaining black box text modules in natural language with language models

   Chandan Singh*, Aliyah R. Hsu*, Richard Antonello, Shailee Jain, Alexander Huth, Bin Yu, and Jianfeng Gao

   In NeurIPS XAI in Action: Past, Present, and Future Applications, 2023

   PDF

2019

1. Personalized Dialogue Response Generation Learned from Monologues

   Feng-Guang Su*, Aliyah R. Hsu*, Yi-Lin Tuan, and Hung-Yi Lee

   In Proc. Interspeech, 2019

   PDF

2018

1. Scan-Mode Atmospheric-Pressure Plasma Jet Processed Reduced Graphene Oxides for Quasi-Solid-State Gel-Electrolyte Supercapacitors

   Aliyah R. Hsu, Hung-Hua Chien, Chen-Yu Liao, Chia-Chun Lee, Jui-Hsuan Tsai, Cheng-Che Hsu, I-Chun Cheng, and Jian-Zhang Chen

   Coatings, 2018

   Abs PDF

   A scanning atmospheric-pressure plasma jet (APPJ) is essential for high-throughput large-area and roll-to-roll processes. In this study, we evaluate scan-mode APPJ for processing reduced graphene oxides (rGOs) that are used as the electrodes of quasi-solid-state gel-electrolyte supercapacitors. rGO nanoflakes are mixed with ethyl cellulose (EC) and terpineol to form pastes for screen-printing. After screen-printing the pastes on carbon cloth, a DC-pulse nitrogen APPJ is used to process the pastes in the scan mode. The maximal temperature attained is  550 °C with a thermal influence duration of  10 s per scan. The pastes are scanned by APPJ for 0, 1, 3 and 5 times. X-ray photoelectron spectroscopy (XPS) indicates the reduction of C-O binding content as the number of scan increases, suggesting the oxidation/decomposition of EC. The areal capacitance increases and then decreases as the number of scan increases; the best achieved areal capacitance is 15.93 mF/cm2 with one APPJ scan, in comparison to 4.38 mF/cm2 without APPJ processing. The capacitance retention rate of the supercapacitor with the best performance is  93% after a 1000-cycle cyclic voltammetry (CV) test. The optimal number of APPJ scans should enable the proper removal of inactive EC and improved wettability while minimizing the damage caused to rGOs by nitrogen APPJ processing.