Search

Your search keyword '"Huang, Haoran"' showing total 457 results
Start Over Author "Huang, Haoran"
457 results on '"Huang, Haoran"'

2. DACO: Towards Application-Driven and Comprehensive Data Analysis via Code Generation

Author

Wu, Xueqing, Zheng, Rui, Sha, Jingzhen, Wu, Te-Lin, Zhou, Hanyu, Tang, Mohan, Chang, Kai-Wei, Peng, Nanyun, and Huang, Haoran

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence
Abstract

Data analysis is a crucial analytical process to generate in-depth studies and conclusive insights to comprehensively answer a given user query for tabular data. In this work, we aim to propose new resources and benchmarks to inspire future research on this crucial yet challenging and under-explored task. However, collecting data analysis annotations curated by experts can be prohibitively expensive. We propose to automatically generate high-quality answer annotations leveraging the code-generation capabilities of LLMs with a multi-turn prompting technique. We construct the DACO dataset, containing (1) 440 databases (of tabular data) collected from real-world scenarios, (2) ~2k query-answer pairs that can serve as weak supervision for model training, and (3) a concentrated but high-quality test set with human refined annotations that serves as our main evaluation benchmark. We train a 6B supervised fine-tuning (SFT) model on DACO dataset, and find that the SFT model learns reasonable data analysis capabilities. To further align the models with human preference, we use reinforcement learning to encourage generating analysis perceived by human as helpful, and design a set of dense rewards to propagate the sparse human preference reward to intermediate code generation steps. Our DACO-RL algorithm is evaluated by human annotators to produce more helpful answers than SFT model in 57.72% cases, validating the effectiveness of our proposed algorithm. Data and code are released at https://github.com/shirley-wu/daco, Comment: NeurIPS 2024 Dataset and Benchmark Track

Published
2024

4. Neuronal BAG3 attenuates tau hyperphosphorylation, synaptic dysfunction, and cognitive deficits induced by traumatic brain injury via the regulation of autophagy-lysosome pathway

Author

Sweeney, Nicholas, Kim, Tae Yeon, Morrison, Cody T., Li, Liangping, Acosta, Diana, Liang, Jiawen, Datla, Nithin V., Fitzgerald, Julie A., Huang, Haoran, Liu, Xianglan, Tan, Gregory Huang, Wu, Min, Karelina, Kate, Bray, Chelsea E., Weil, Zachary M., Scharre, Douglas W., Serrano, Geidy E., Saito, Takashi, Saido, Takaomi C., Beach, Thomas G., Kokiko-Cochran, Olga N., Godbout, Jonathan P., Johnson, Gail V. W., and Fu, Hongjun

Published
2024
Full Text
View/download PDF

7. Improving Generalization of Alignment with Human Preferences through Group Invariant Learning

Author

Zheng, Rui, Shen, Wei, Hua, Yuan, Lai, Wenbin, Dou, Shihan, Zhou, Yuhao, Xi, Zhiheng, Wang, Xiao, Huang, Haoran, Gui, Tao, Zhang, Qi, and Huang, Xuanjing

Subjects
Computer Science - Machine Learning, Computer Science - Artificial Intelligence
Abstract

The success of AI assistants based on language models (LLMs) hinges crucially on Reinforcement Learning from Human Feedback (RLHF), which enables the generation of responses more aligned with human preferences. As universal AI assistants, there's a growing expectation for them to perform consistently across various domains. However, previous work shows that Reinforcement Learning (RL) often exploits shortcuts to attain high rewards and overlooks challenging samples. This focus on quick reward gains undermines both the stability in training and the model's ability to generalize to new, unseen data. In this work, we propose a novel approach that can learn a consistent policy via RL across various data groups or domains. Given the challenges associated with acquiring group annotations, our method automatically classifies data into different groups, deliberately maximizing performance variance. Then, we optimize the policy to perform well on challenging groups. Lastly, leveraging the established groups, our approach adaptively adjusts the exploration space, allocating more learning capacity to more challenging data and preventing the model from over-optimizing on simpler data. Experimental results indicate that our approach significantly enhances training stability and model generalization.

Published
2023

8. Secrets of RLHF in Large Language Models Part I: PPO

Author

Zheng, Rui, Dou, Shihan, Gao, Songyang, Hua, Yuan, Shen, Wei, Wang, Binghai, Liu, Yan, Jin, Senjie, Liu, Qin, Zhou, Yuhao, Xiong, Limao, Chen, Lu, Xi, Zhiheng, Xu, Nuo, Lai, Wenbin, Zhu, Minghao, Chang, Cheng, Yin, Zhangyue, Weng, Rongxiang, Cheng, Wensen, Huang, Haoran, Sun, Tianxiang, Yan, Hang, Gui, Tao, Zhang, Qi, Qiu, Xipeng, and Huang, Xuanjing

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence, Computer Science - Machine Learning
Abstract

Large language models (LLMs) have formulated a blueprint for the advancement of artificial general intelligence. Its primary objective is to function as a human-centric (helpful, honest, and harmless) assistant. Alignment with humans assumes paramount significance, and reinforcement learning with human feedback (RLHF) emerges as the pivotal technological paradigm underpinning this pursuit. Current technical routes usually include \textbf{reward models} to measure human preferences, \textbf{Proximal Policy Optimization} (PPO) to optimize policy model outputs, and \textbf{process supervision} to improve step-by-step reasoning capabilities. However, due to the challenges of reward design, environment interaction, and agent training, coupled with huge trial and error cost of large language models, there is a significant barrier for AI researchers to motivate the development of technical alignment and safe landing of LLMs. The stable training of RLHF has still been a puzzle. In the first report, we dissect the framework of RLHF, re-evaluate the inner workings of PPO, and explore how the parts comprising PPO algorithms impact policy agent training. We identify policy constraints being the key factor for the effective implementation of the PPO algorithm. Therefore, we explore the PPO-max, an advanced version of PPO algorithm, to efficiently improve the training stability of the policy model. Based on our main results, we perform a comprehensive analysis of RLHF abilities compared with SFT models and ChatGPT. The absence of open-source implementations has posed significant challenges to the investigation of LLMs alignment. Therefore, we are eager to release technical reports, reward models and PPO codes, aiming to make modest contributions to the advancement of LLMs.

Published
2023

9. Calibration Set Optimization by Dragonfly Algorithm for Near-Infrared Modeling of Wheat Flour Protein Content

Author

HU Yunchao, LIU Zhijian, WANG Ying, HUANG Haoran, WANG Honghong, WU Cai’e, XIONG Zhixin

Subjects
dragonfly algorithm, near-infrared spectroscopy, optimization of calibration set, protein content of wheat flour, Food processing and manufacture, TP368-456
Abstract

In order to optimize the calibration set for near-infrared modeling of the protein content in wheat flour, the binary dragonfly algorithm (BDA) was used to select representative samples from the primary calibration set divided by the traditional Kennard/Stone (K/S) method. Based on the representative samples, a partial least square regression (PLSR) model for estimating the protein content in wheat flour was established, and the prediction set was employed to evaluate the stability and prediction performance of the model. The results indicated that an optimal calibration set with 30 samples was selected finally by BDA, and the proposed model exhibited a coefficient of determination of prediction (Rp2) of 0.956 4 and a root mean square errors of prediction (RMSEP) of 0.278 1, which increased by 1.87% and decreased by 15.57% compared with those (0.938 8 and 0.329 4) from K/S partition of 100 primary calibration sets, respectively. The average number of calibration sets selected from 10 BDA experiments was 30.2, and the protein content of wheat flour was predicted better by the 10 models developed than that obtained based on the primary calibration set. Therefore, BDA can select a small number of representative calibration set samples based on which a PLSR model with good robustness and high prediction accuracy for the protein content of wheat flour can be established. The proposed method can provide an efficient tool for calibration set selection in near-infrared spectroscopic analysis of the quality of wheat flour.

Published
2024
Full Text
View/download PDF

11. Research on Cost Control of Mobile Project Procurement Based on Improved SEIRS Model

Author

Liu, Hongyu, Huang, Haoran, Chan, Albert P. C., Series Editor, Hong, Wei-Chiang, Series Editor, Mellal, Mohamed Arezki, Series Editor, Narayanan, Ramadas, Series Editor, Nguyen, Quang Ngoc, Series Editor, Ong, Hwai Chyuan, Series Editor, Sachsenmeier, Peter, Series Editor, Sun, Zaicheng, Series Editor, Ullah, Sharif, Series Editor, Wu, Junwei, Series Editor, Zhang, Wei, Series Editor, Chen, Colin W. K., editor, Malik, Tariq H., editor, Fu, Qiufang, editor, and Xuan, Haiyan, editor

Published
2024
Full Text
View/download PDF

17. Two-Level Supervised Contrastive Learning for Response Selection in Multi-Turn Dialogue

Author

Zhang, Wentao, Xu, Shuang, and Huang, Haoran

Subjects
Computer Science - Computation and Language
Abstract

Selecting an appropriate response from many candidates given the utterances in a multi-turn dialogue is the key problem for a retrieval-based dialogue system. Existing work formalizes the task as matching between the utterances and a candidate and uses the cross-entropy loss in learning of the model. This paper applies contrastive learning to the problem by using the supervised contrastive loss. In this way, the learned representations of positive examples and representations of negative examples can be more distantly separated in the embedding space, and the performance of matching can be enhanced. We further develop a new method for supervised contrastive learning, referred to as two-level supervised contrastive learning, and employ the method in response selection in multi-turn dialogue. Our method exploits two techniques: sentence token shuffling (STS) and sentence re-ordering (SR) for supervised contrastive learning. Experimental results on three benchmark datasets demonstrate that the proposed method significantly outperforms the contrastive learning baseline and the state-of-the-art methods for the task.

Published
2022

20. Discovery of novel activators of large-conductance calcium-activated potassium channels for the treatment of cerebellar ataxia

Author

Srinivasan, Sharan R, Huang, Haoran, Chang, Wei-Chih, Nasburg, Joshua A, Nguyen, Hai M, Strassmaier, Tim, Wulff, Heike, and Shakkottai, Vikram G

Subjects
Medical Physiology, Biomedical and Clinical Sciences, Orphan Drug, Neurodegenerative, Brain Disorders, Rare Diseases, Neurosciences, 5.1 Pharmaceuticals, 1.1 Normal biological development and functioning, Neurological, Animals, Ataxia, Calcium, Cerebellar Ataxia, Large-Conductance Calcium-Activated Potassium Channels, Mice, Potassium, Potassium Channels, Calcium-Activated, Spinocerebellar Ataxias, Biochemistry and Cell Biology, Pharmacology and Pharmaceutical Sciences, Pharmacology & Pharmacy, Biochemistry and cell biology, Pharmacology and pharmaceutical sciences
Abstract

Impaired cerebellar Purkinje neuron firing resulting from reduced expression of large-conductance calcium-activated potassium (BK) channels is a consistent feature in models of inherited neurodegenerative spinocerebellar ataxia (SCA). Restoring BK channel expression improves motor function and delays cerebellar degeneration, indicating that BK channels are an attractive therapeutic target. Current BK channel activators lack specificity and potency and are therefore poor templates for future drug development. We implemented an automated patch clamp platform for high-throughput drug discovery of BK channel activators using the Nanion SyncroPatch 384PE system. We screened over 15,000 compounds for their ability to increase BK channel current amplitude under conditions of lower intracellular calcium that is present in disease. We identified several novel BK channel activators that were then retested on the SyncroPatch 384PE to generate concentration-response curves (CRCs). Compounds with favorable CRCs were subsequently tested for their ability to improve irregular cerebellar Purkinje neuron spiking, characteristic of BK channel dysfunction in SCA1 mice. We identified a novel BK channel activator, 4-chloro-N-(5-chloro-2-cyanophenyl)-3-(trifluoromethyl)benzene-1-sulfonamide (herein renamed BK-20), that exhibited a more potent half-maximal activation of BK current (pAC50 = 4.64) than NS-1619 (pAC50 = 3.7) at a free internal calcium concentration of 270 nM in a heterologous expression system and improved spiking regularity in SCA1 Purkinje neurons. BK-20 had no activity on small-conductance calcium-activated potassium (SK)1-3 channels but interestingly was a potent blocker of the T-type calcium channel, Cav3.1 (IC50 = 1.05 μM). Our work describes both a novel compound for further drug development in disorders with irregular Purkinje spiking and a unique platform for drug discovery in degenerative ataxias. SIGNIFICANCE STATEMENT: Motor impairment associated with altered Purkinje cell spiking due to dysregulation of large-conductance calcium-activated potassium (BK) channel expression and function is a shared feature of disease in many degenerative ataxias. BK channel activators represent an outstanding therapeutic agent for ataxia. We have developed a high-throughput platform to screen for BK channel activators and identified a novel compound that can serve as a template for future drug development for the treatment of these disabling disorders.

Published
2022

26. Spelling Error Correction with Soft-Masked BERT

Author

Zhang, Shaohua, Huang, Haoran, Liu, Jicong, and Li, Hang

Subjects
Computer Science - Computation and Language, Computer Science - Machine Learning
Abstract

Spelling error correction is an important yet challenging task because a satisfactory solution of it essentially needs human-level language understanding ability. Without loss of generality we consider Chinese spelling error correction (CSC) in this paper. A state-of-the-art method for the task selects a character from a list of candidates for correction (including non-correction) at each position of the sentence on the basis of BERT, the language representation model. The accuracy of the method can be sub-optimal, however, because BERT does not have sufficient capability to detect whether there is an error at each position, apparently due to the way of pre-training it using mask language modeling. In this work, we propose a novel neural architecture to address the aforementioned issue, which consists of a network for error detection and a network for error correction based on BERT, with the former being connected to the latter with what we call soft-masking technique. Our method of using `Soft-Masked BERT' is general, and it may be employed in other language detection-correction problems. Experimental results on two datasets demonstrate that the performance of our proposed method is significantly better than the baselines including the one solely based on BERT., Comment: To be published at ACL 2020

Published
2020

48. Characteristic volatile compounds of white tea with different storage times using E‐nose, HS–GC–IMS, and HS–SPME–GC–MS.

Author

Huang, Haoran, Chen, Xinyu, Wang, Ying, Cheng, Ye, Liu, Zhijian, Hu, Yunchao, Wu, Xianzhi, Wu, Caie, and Xiong, Zhixin

Subjects
*PROPIONIC acid, *ELECTRONIC noses, *QUALITY control, *ANALYSIS of variance, *SENSORY evaluation, *ION mobility spectroscopy
Abstract

This paper studied the influence of storage duration on the flavor profile of white tea in detail, with samples produced between 2020 and 2023. Sensory evaluation was performed by quantitative descriptive analysis (QDA), followed by an in‐depth aroma components analysis employing an electronic nose (E‐nose), headspace gas chromatography–ion mobility spectrometry (HS–GC–IMS), and headspace solid–phase microextraction–gas chromatography–mass spectrometry (HS–SPME–GC–MS). The QDA findings revealed a gradual transition in the flavor profile of white tea during storage, shifting from sweet, fruity, and floral to more herbal and stale characteristics. E‐nose could well distinguish white tea with different storage times. A total of 55 and 53 volatile compounds were identified by HS–GC–IMS and HS–SPME–GC–MS, respectively. The orthogonal partial least squares‐discriminant analysis models, based on HS–GC–IMS (R2Y = 0.998, Q2 = 0.987) and HS–SPME–GC–MS (R2Y = 0.984, Q2 = 0.993), successfully distinguished white tea samples stored for different storage times. Furthermore, 14 and 8 key compounds were screened based on the double variable criterion of one‐way analysis of variance (p < 0.05) and variable importance in projection (VIP) >1.2, and their content changes were also compared. It is the gradual decrease of important aroma components such as 2‐hexenal, 2‐methyl‐2‐hepten‐6‐one, linalool, and geraniol, which are positively correlated with sweet, fruity, and floral aromas, and the gradual increase of hexanoic acid, thiophene, propanoic acid, dimethyl disulfide, and borneyl acetate, which are positively correlated with herbal and stale flavor, that leads to the changes in flavor and aroma of white tea during storage. The results of the study provided a reference for elucidating the aroma characteristics of white tea at different storage times as well as a theoretical basis for the quality control of white tea. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

49. Construction and Validation of a Novel T/NK-Cell Prognostic Signature for Pancreatic Cancer Based on Single-Cell RNA Sequencing.

Author

Wang, Yu, Zhang, Cong, Zhang, Jianlu, Huang, Haoran, and Guo, Junchao

Subjects
KILLER cells, PREDICTION models, RESEARCH funding, RESEARCH methodology evaluation, DESCRIPTIVE statistics, LYMPHOCYTES, TUMOR markers, PANCREATIC tumors, METASTASIS, BIOINFORMATICS, RNA, GENE expression profiling, SEQUENCE analysis
Abstract

Background: Evidence with regards to the distinction between primary and metastatic tumors in pancreatic cancer and driving factors for metastases remains limited. Methods: Single-cell RNA sequencing (scRNA-seq) was conducted on metastatic pancreatic cancer. Bioinformatics analysis on relevant sequencing data was used to construct a risk model to predict patient prognosis. Furthermore, immune infiltration and metabolic differences were assessed. The biological function of key differential genes was evaluated. Results: Paired primary and metastatic tumor tissues from 3 pancreatic cancer patients were collected and conducted scRNA-seq. Subsequently, the T/NK cell subgroup was the most different cell type between primary tumors and liver metastases and was selected for further analysis. Eventually, 6 specifically expressed genes of T/NK cells (B2M, ZFP36L2, ANXA1, ARL4C, TSPYL2, FYN) were used constructing the prognostic model. The stability of this model was validated by an external cohort. Meanwhile, different immune infiltration abundances occurred between high and low risk groups stratified by the model. The high-risk group had a stronger metabolic capability. Conclusions: A novel prognostic T/NK-cell signature for pancreatic cancer was constructed based on scRNA-seq data and externally validated. The involved key genes may play a role in multiple metabolic pathways of metastasis and affect the tumor immune microenvironment. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

50. Energy-Efficiency-Oriented Spatial Configuration of VRV Outdoor Units in an Equipment Layer Under Background Wind Conditions.

Author

Liu, Lin, Huang, Haoran, and Tian, Xiaoyu

Subjects
WIND speed, ATMOSPHERIC temperature, TEMPERATURE distribution, INLETS, COOLING
Abstract

This study provides a spatial configuration method to improve the cooling efficiency of multiple VRV outdoor units placed on equipment layers with high floors. Relevant factors include wind parameters, the placement of multiple outdoor units, and louver. A total of 96 cases were designed. CFD simulations were used to obtain the inlet air temperature distributions of multiple outdoor units and then calculate their cooling efficiency. The results found that these factors have effects on the average cooling efficiency of outdoor units in a single row to a certain extent. The influencing degrees of these factors, from large to small, were the louver angle, wind parameters, and the placement of multiple outdoor units. When the cooling efficiency of outdoor units was maximum and the louver angle was 15°, the louvers could be oriented perpendicular to the dominant wind direction (90°) when wind speed was ≥6 m/s, and the unit spacing was 600 mm. Based on this, when the number of outdoor units was expanded in the limited space, staggered arrangements with different directions of heat exchange surfaces were a recommended optimization layout. This study provides technical support for improving the working efficiency of VRV outdoor units in an equipment layer. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results