Junyi Ye

Junyi Ye

Assistant Professor at Montclair State University. Research: LLMs, graph neural networks, time series, computer vision.

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

portfolio

Coarse2Fine ResNet

A robust and high-precision generalized deep learning framework for time delay estimation

CreativeMath

Assessing the creativity of LLMs in proposing novel solutions to mathematical problems

DataFrame QA

A universal LLM framework on DataFrame question answering without data exposure

DySTAGE

Dynamic graph representation learning for asset pricing via spatio-temporal attention and graph encodings

FaultyMath

Benchmarking LLMs’ logical integrity on faulty mathematical problems

Margin Trader LLM

Adaptive and explainable margin trading via LLMs on portfolio management

SafeLight

A reinforcement learning method toward collision-free traffic signal control

Solar GAN

High resolution solar image generation using generative adversarial networks

TextFlow

Leveraging intermediate text representations for superior flowchart understanding

publications

Prediction with Time-Series Mixer for the S&P500 Index

Published in 2023 IEEE 39th International Conference on Data Engineering Workshops (ICDEW), 20–27, 2023

A Time-Series Mixer (TS-Mixer) architecture, based on MLP-Mixer, for multivariate time series forecasting applied to S&P500 Index prediction.

Recommended citation: Ye, J., Gu, J., Dash, A., Deek, F. P., & Wang, G. (2023). "Prediction with Time-Series Mixer for the S&P500 Index." 2023 IEEE 39th International Conference on Data Engineering Workshops (ICDEW), 20–27.

SafeLight: A Reinforcement Learning Method toward Collision-Free Traffic Signal Control

Published in Proceedings of the AAAI Conference on Artificial Intelligence, 37(12), 14801–14810, 2023

A safety-enhanced residual reinforcement learning method for traffic signal control that reduces collisions while increasing mobility.

Recommended citation: Du, W., Ye, J., Gu, J., Li, J., Wei, H., & Wang, G. (2023). "SafeLight: A Reinforcement Learning Method toward Collision-Free Traffic Signal Control." Proceedings of the AAAI Conference on Artificial Intelligence, 37(12), 14801–14810.
Download Paper

Establishing a Baseline for Evaluating Blockchain-Based Self-Sovereign Identity Systems: A Systematic Approach to Assess Capability, Compatibility and Interoperability

Published in Proceedings of the 2024 6th Blockchain and Internet of Things Conference (BIOTC), 108–119, 2024

A systematic approach and baseline for evaluating capability, compatibility, and interoperability of blockchain-based self-sovereign identity (SSI) systems.

Recommended citation: Yao, W., Du, W., Gu, J., Ye, J., Deek, F. P., & Wang, G. (2024). "Establishing a Baseline for Evaluating Blockchain-Based Self-Sovereign Identity Systems." Proceedings of the 2024 6th Blockchain and Internet of Things Conference, 108–119.

A Review of Generative Adversarial Networks (GANs) and Its Applications in a Wide Variety of Disciplines: From Medical to Remote Sensing

Published in IEEE Access, 12, 18330–18357, 2024

A comprehensive survey of GAN theory, variants, evaluation metrics, and applications across twelve domains.

Recommended citation: Dash, A., Ye, J., & Wang, G. (2024). "A Review of Generative Adversarial Networks (GANs) and Its Applications in a Wide Variety of Disciplines: From Medical to Remote Sensing." IEEE Access, 12, 18330–18357.

High Resolution Solar Image Generation Using Generative Adversarial Networks

Published in Annals of Data Science, 11(5), 1545–1561, 2024

Using GANs (Pix2Pix / Pix2PixHD) to translate SDO/HMI magnetogram images into high-resolution SDO/AIA 0304-Ã… images.

Recommended citation: Dash, A., Ye, J., Wang, G., & Jin, H. (2024). "High Resolution Solar Image Generation Using Generative Adversarial Networks." Annals of Data Science, 11(5), 1545–1561.
Download Paper

From Blind Solvers to Logical Thinkers: Benchmarking LLMs’ Logical Integrity on Faulty Mathematical Problems

Published in arXiv preprint arXiv:2410.18921, 2024

FaultyMath is a benchmark of logically flawed math problems used to test whether LLMs merely calculate or actually reason about problem validity.

Recommended citation: Rahman*, M., Ye*, J., Yao, W., Yin, W., & Wang, G. (2024). "From Blind Solvers to Logical Thinkers: Benchmarking LLMs' Logical Integrity on Faulty Mathematical Problems." arXiv preprint arXiv:2410.18921.
Download Paper

DySTAGE: Dynamic Graph Representation Learning for Asset Pricing via Spatio-Temporal Attention and Graph Encodings

Published in 5th ACM International Conference on AI in Finance (ICAIF 2024), 2024

A dynamic graph representation learning framework for asset pricing that adapts to changing asset pools and correlations over time.

Recommended citation: Gu*, J., Ye*, J., Uddin, A., & Wang, G. (2024). "DySTAGE: Dynamic Graph Representation Learning for Asset Pricing via Spatio-Temporal Attention and Graph Encodings." Proceedings of the 5th ACM International Conference on AI in Finance (ICAIF '24), 388–396.
Download Paper

Adaptive and Explainable Margin Trading via Large Language Models on Portfolio Management

Published in 5th ACM International Conference on AI in Finance (ICAIF 2024), 2024

An adaptive, explainable framework combining LLMs and reinforcement learning for dynamic long-short position adjustment.

Recommended citation: Gu*, J., Ye*, J., Wang, G., & Yin, W. (2024). "Adaptive and Explainable Margin Trading via Large Language Models on Portfolio Management." Proceedings of the 5th ACM International Conference on AI in Finance (ICAIF '24), 248–256.
Download Paper

Assessing the Creativity of LLMs in Proposing Novel Solutions to Mathematical Problems

Published in 39th AAAI Conference on Artificial Intelligence (AAAI 2025), 2025

CreativeMath is a benchmark for assessing whether LLMs can propose novel solutions to math problems after being shown known solutions.

Recommended citation: Ye, J., Gu, J., Zhao, X., Yin, W., & Wang, G. (2025). "Assessing the Creativity of LLMs in Proposing Novel Solutions to Mathematical Problems." Proceedings of the AAAI Conference on Artificial Intelligence.
Download Paper

Beyond End-to-End VLMs: Leveraging Intermediate Text Representations for Superior Flowchart Understanding

Published in 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics (NAACL 2025), 2025

TextFlow decomposes flowchart understanding into a Vision Textualizer and a Textual Reasoner, improving controllability and explainability over end-to-end VLMs.

Recommended citation: Ye, J., Dash, A., Yin, W., & Wang, G. (2025). "Beyond End-to-End VLMs: Leveraging Intermediate Text Representations for Superior Flowchart Understanding." Proceedings of NAACL 2025.
Download Paper

Illusions in Humans and AI: How Visual Perception Aligns and Diverges

Published in arXiv preprint arXiv:2508.12422, 2025

A comparison of human and AI visual illusions, uncovering alignment gaps and AI-specific perceptual vulnerabilities absent in human perception.

Recommended citation: Yang*, J., Ye*, J., Dash, A., & Wang, G. (2025). "Illusions in Humans and AI: How Visual Perception Aligns and Diverges." arXiv preprint arXiv:2508.12422.
Download Paper

Calibration Bets on the Past: Post-Training Quantization for Financial Time-Series Forecasting

Published in 7th ACM International Conference on AI in Finance (ICAIF 2026), 2026

A systematic study showing that activation calibration is the primary determinant of predictive performance for 4-bit post-training quantization in financial forecasting.

Recommended citation: Ye, J., & Wanjiku, I. G. (2026). "Calibration Bets on the Past: Post-Training Quantization for Financial Time-Series Forecasting." Proceedings of the 7th ACM International Conference on AI in Finance (ICAIF '26).
Download Paper

Regime-Gated Residual Mixture-of-Experts for Cross-Sectional Volatility Forecasting

Published in 7th ACM International Conference on AI in Finance (ICAIF 2026), 2026

RG-ResMoE routes regime information through expert gating rather than direct forecasting, improving cross-sectional volatility forecasting accuracy and training stability.

Recommended citation: Ye, J., & Borde, G. V. (2026). "Regime-Gated Residual Mixture-of-Experts for Cross-Sectional Volatility Forecasting." Proceedings of the 7th ACM International Conference on AI in Finance (ICAIF '26).
Download Paper

Damage Predicts Recovery: When Calibration Data Matters in Compressing Financial LLMs

Published in 7th ACM International Conference on AI in Finance (ICAIF 2026), 2026

A study across two model families and ten financial tasks showing that calibration-corpus choice for compressing LLMs only matters when compression causes large task-specific damage.

Recommended citation: Ye, J., Yu, M., Hazra, D., & Wang, G. (2026). "Damage Predicts Recovery: When Calibration Data Matters in Compressing Financial LLMs." Proceedings of the 7th ACM International Conference on AI in Finance (ICAIF '26).
Download Paper

Can You Delete a Year of Market Data? Machine Unlearning Against Exact Retraining Oracles

Published in 7th ACM International Conference on AI in Finance (ICAIF 2026), 2026

A 3,200-model benchmark of temporal machine unlearning in financial forecasting against exact retraining oracles, showing when approximate deletion methods can close the reference-to-oracle gap.

Recommended citation: Ye, J. (2026). "Can You Delete a Year of Market Data? Machine Unlearning Against Exact Retraining Oracles." Proceedings of the 7th ACM International Conference on AI in Finance (ICAIF '26).
Download Paper

teaching

Database Systems

CSIT 355, Montclair State University

Database Systems

CSIT 355, Montclair State University

Advanced Techniques in Data Science

CSIT 360, Montclair State University

Database Systems

CSIT 555, Montclair State University

Advanced Techniques in Data Science

CSIT 557, Montclair State University

Data Structures and Algorithms

CSIT 212, Montclair State University

Database Systems

CSIT 355, Montclair State University