📄 Selected Publications

I have published and presented my work at prestigious conferences and journals, including ICML, Journal of Marine Science and Engineering, Sensors, IEEE conferences, and IISE, as well as workshops at various international venues. A curated selection is listed below, organized by year (newest first).

2026

2026 Position: Time-Series Foundation Models Require Explicit Domain-Level Benchmarks · Syed, M. A. B., Ahamed, M. Y., & Wasi, A. T. (2026) · ICML 2026 · OpenReview · ICML

2026 Towards Efficient Real-Time Video Motion Transfer via Generative Time Series Modeling · Haque, T., Syed, M. A. B., Jeong, B., Bai, X., Mohan, S., Paul, S., Ahmed, I., & Das, S. (2026) · Multimedia Tools and Applications (Springer) · Springer · DOI
Show Abstract

Abstract: This paper presents a novel approach for real-time video motion transfer using generative time series modeling. We propose an efficient framework that transfers motion patterns from source videos to target subjects while maintaining temporal consistency and visual quality. Our method leverages generative models to learn motion representations and enables real-time processing for applications in video editing, animation, and virtual content creation.

2025

2025 Zero-Shot Time-Series Forecasting: Do Time-Series FMs Outperform Domain-Agnostic FMs? · Syed, M. A. B., et al. (2025) · NeurIPS 2025 Workshop (BERT²S) · OpenReview · NeurIPS
Show Abstract

Abstract: Foundation models (FMs) have achieved major advances in language, vision, and speech. In parallel, time-series foundation models (TSFMs) have been developed to address forecasting tasks. A key question is whether TSFMs truly generalize to unseen time series data, and whether they perform better than general-purpose FMs from other domains in a zero-shot setting. We compare four TSFMs such as Chronos, Times-FM, TimeGPT, and MOMENTs with cross-domain FMs for text (GPT), audio (Whisper), and vision (ViT). For a systematic comparison, we use simple task-agnostic adapters to convert sequences into forecasts, without fine-tuning or changing the backbone models. All models are evaluated on nine diverse datasets that were unseen during training. Our results show that TSFMs perform best on most datasets, highlighting the benefit of temporal pretraining and time-aware design. Overall, the strong zero-shot performance of TSFMs suggests that they may represent a breakthrough comparable to BERT for time series forecasting. At the same time, large text-based models such as GPT remain surprisingly competitive, in some cases even surpassing TSFMs, highlighting the ability of general-purpose models to capture temporal patterns despite not being trained for this task. GitHub repository: https://github.com/anonymous4865/tsfms.

2025 DIVA: Diversity Assessment in Text-to-Image Generation via Hybrid Metrics · Syed, M. A. B., Ahamed, M. Y. (2025) · ICML 2025 Workshop (4th Muslims in ML Workshop) · OpenReview · ICML
Show Abstract

Abstract: Generative models like Stable Diffusion, DALL-E, and Imagen have shown impressive capabilities in creating visually compelling images from textual prompts. However, not all models produce a wide variety of outputs from the same prompt. In some applications—such as creative advertising or artistic design—diverse outputs are highly valued for exploring different visual interpretations. In contrast, tasks like forensic analysis or technical illustration require high consistency to ensure reproducibility. Current diversity quantification methods, such as Bayesian frameworks and pixel-based metrics (e.g., FID, SSIM), either ignore prompt-specific variability or fail to disentangle aleatoric and epistemic factors. In this work, we present DIVA, a framework quantifying diversity through hybrid diversity metrics: mean pairwise CLIP embedding distance, feature distribution variance, and information entropy. DIVA integrates these metrics into a unified diversity score, capturing both aleatoric and epistemic uncertainty. It adapts to both diversity-expected prompts and diversity-constrained prompts. Human validation shows strong correlation between our diversity score and human judgments. This work provides a scalable solution for applications requiring reliability and transparency, from creative design to medical imaging. GitHub repository: https://github.com/anonymous4865/diva.

2025 A systematic review of time series algorithms and analytics in predictive maintenance · Syed, M. A. B., Hasan, M. R., Chowdhury, N. I., Rahman, M. H., & Ahmed, I. (2025) · Decision Analytics Journal, 100573 · DOI
Show Abstract

Abstract: This systematic review provides a comprehensive analysis of time series algorithms and analytics in predictive maintenance applications. We examine various machine learning and deep learning approaches for forecasting equipment failures, analyzing sensor data patterns, and optimizing maintenance schedules. The review covers recent advances in time series forecasting, anomaly detection, and remaining useful life prediction, offering insights into the current state-of-the-art and future research directions in Industry 4.0 and Industry 5.0 contexts.

2024

2024 Understanding AI Chatbot adoption in education: PLS-SEM analysis of user behavior factors · Hasan, M. R., Chowdhury, N. I., Rahman, M. H., Syed, M. A. B., & Ryu, J. H. (2024) · Computers in Human Behavior: Artificial Humans, 2(2), 100098 · DOI
Show Abstract

Abstract: This research employs Partial Least Squares Structural Equation Modeling (PLS-SEM) to investigate the factors influencing AI chatbot adoption in educational settings. We examine user behavior patterns, technology acceptance factors, and educational outcomes associated with chatbot implementation. Our findings reveal key determinants of adoption including perceived usefulness, ease of use, and educational value, providing insights for educators and technology developers seeking to enhance learning experiences through AI-powered conversational interfaces.

2024 Predictive health analysis in industry 5.0: A scientometric and systematic review of Motion Capture in construction · Rahman, M. H., Hasan, M. R., Chowdhury, N. I., Syed, M. A. B., & Farah, M. U. (2024) · Digital Engineering, 1, 100002 · DOI
Show Abstract

Abstract: This scientometric and systematic review examines the application of motion capture technology in construction industry health and safety analysis within the Industry 5.0 framework. We analyze research trends, technological implementations, and safety outcomes associated with motion capture systems for worker monitoring, ergonomic assessment, and injury prevention in construction environments.

View All Publications →

🛠️ Technical Skills

• Programming Languages: Python, R, SQL, C
• ML/DL Frameworks: Scikit-learn, Keras, TensorFlow, PyTorch
• Data Analysis: MS Excel, Tableau, Power BI, Minitab
• Cloud Platforms: AWS, Azure ML, GCP, Vertex AI
• MLOps & DevOps: Docker, Kubernetes, Jenkins, Git, GitHub Actions, MLflow, DVC, Weights & Biases, nginx
• Distributed ML & HPC: Ray, SLURM, HPC clusters
• Databases: BigQuery, MySQL, PostgreSQL
• Model Deployment: FastAPI, Flask, vLLM, TensorFlow Serving, Model Monitoring, A/B Testing, CI/CD Pipelines
• Other Tools: SAP Ariba, SAP MDCS, SolidWorks, MS Project, MS Access, CPLEX, HTML, CSS

🔬 Research Interest

Most of my work lately falls into two areas: time-series foundation models, and agentic systems built on top of large language models.

On the time-series side, I care about whether these models actually hold up when you split benchmarks by domain, how they perform zero-shot on new datasets, and how generative time-series methods can support things like real-time video motion transfer. In industry, I’ve also spent a lot of time on offline reinforcement learning and forecasting for supply chain problems at The Home Depot.

For language models, I’m working on reasoning models, process reward models (PRMs), tool calling, and agentic AI security—the practical question is how to get reliable multi-step behavior without something breaking once you put the system in front of real users. One example is Adhawk, a human-in-the-loop tool we built to curate customer lists using natural language and SQL. I also study how to measure diversity in generative image models, and I still publish on physics-informed deep learning for marine surveillance when the problem calls for it. If you’re working on any of these topics and think there’s overlap, I’m happy to collaborate—get in touch.

📰 News and Updates

Consistency is the key to self-improvement, and my GitHub activity serves as a powerful reminder of my commitment to continuous learning and coding progress.