Parisa Kavianpour

Applied Machine Learning Researcher

Deep Learning for Time-Series · Graphs · Language

I'm an applied machine-learning researcher with an M.Sc. in Earthquake Engineering and several years of experience building deep-learning systems across both scientific and industrial settings. My work moves between three connected areas: spatiotemporal time-series forecasting, graph neural networks with domain adaptation for industrial fault diagnosis, and natural language processing with large language models (RAG, fine-tuning, information extraction).

A recurring theme across my research is making models work under hard, real-world conditions — incomplete data, weak labels, distribution shift, and limited compute. I've published this work in highly cited venues, deployed it in production, and reviewed it for several Elsevier and IEEE journals.

I'm currently seeking a PhD position in applied AI / machine learning in Europe, where I can take this work further on problems with real scientific or societal impact.

• Deep Learning for Time-Series & Spatiotemporal Data — sequence models (CNN-BiLSTM, attention) for forecasting in noisy, chaotic, real-world signals.
• Graph Neural Networks & Domain Adaptation — transferable, data-efficient models for reliability assessment and fault diagnosis under distribution shift.
• NLP & Large Language Models — retrieval-augmented generation, parameter-efficient fine-tuning (LoRA/QLoRA), information extraction, and recommender systems.
• Trustworthy & Data-Efficient ML — learning from incomplete, weakly-labeled, and imbalanced data; physics-informed and knowledge-distilled models.

Full list on Google Scholar · ResearchGate. 279 citations · h-index 5 · i10-index 4.

1. A CNN-BiLSTM Model with Attention Mechanism for Earthquake Prediction. P. Kavianpour, M. Kavianpour, E. Jahani, A. Ramezani. The Journal of Supercomputing (Springer), 2023. — 200+ citations · code

2. Knowledge Distillation and Enhanced Subdomain Adaptation Using Graph Convolutional Network for Resource-Constrained Fault Diagnosis. M. Kavianpour, P. Kavianpour, A. Ramezani, M. Beheshti. Knowledge-Based Systems (Elsevier), 2025. Q1 · IF 7.6 · code

3. A Partial-Imbalance Robust Domain Adaptation Framework for Bearing Fault Diagnosis Using Physics-Informed Deep Learning. M. Kavianpour, P. Kavianpour, A. Ramezani. Measurement (Elsevier), 2025. Q1 · IF 5.6 · code

4. Deep Multi-scale Dilated Convolution Neural Network with Attention Mechanism: A Novel Method for Earthquake Magnitude Classification. P. Kavianpour, M. Kavianpour, A. Ramezani. ICSPIS (IEEE), 2022.

5. Earthquake Magnitude Prediction Using Spatio-temporal Features Learning Based on a Hybrid CNN-BiLSTM Model. P. Kavianpour, M. Kavianpour, E. Jahani, A. Ramezani. ICSPIS (IEEE), 2021.

6. An Intelligent Gearbox Fault Diagnosis under Different Operating Conditions Using Adversarial Domain Adaptation. M. Kavianpour, M. Ghorvei, P. Kavianpour, A. Ramezani, M. T. H. Beheshti. ICCIA (IEEE), 2022.

Languages

Deep Learning & ML

LLMs & NLP

Methods — CNNs · RNNs/LSTM · Transformers · Attention · GNNs · Autoencoders · Domain Adaptation · Transfer Learning · Recommender Systems

Engineering — FastAPI · Flask · Django · Docker · Git · AWS · GCP · Azure · MongoDB · MySQL · Redis · Linux

Peer Reviewer for Measurement (Elsevier), Expert Systems with Applications (Elsevier), Engineering Applications of Artificial Intelligence (Elsevier), IEEE Transactions on Industrial Electronics, and IEEE Transactions on Instrumentation and Measurement.