A full list of publications is available on Google Scholar. For access to PDFs, please get in touch with me via ResearchGate or email.
We are exploring this field along two main directions. First, we focus on developing ML algorithms that can run on quantum hardware. Because quantum computing is fundamentally different from classical computing, we adapt existing algorithms to the quantum paradigm. For example, we recently worked with 1D quantum convolution, also known as quanvolution [1]. Second, we take a more conceptual approach by designing ML models inspired (either entirely or partially) by quantum mechanics. The idea is that such models may offer advantages over classical approaches in terms of generalizability, robustness, and parameter efficiency. Our systematic review [2] and investigations across several domains [1, 3], including mental health [4], highlight these potential benefits and point to the promise of quantum ML. We are also exploring how quantum ML can open up new possibilities for neural representation learning [5].
From a data science perspective, the broad field of neuroinformatics can be divided into several modalities: signals (e.g., EEG, MEG), images (e.g., MRI, CT), tabular data (e.g., cognitive assessments), texts (e.g., EHRs, clinical notes), and omics (e.g., genomics, proteomics). Our goal is to integrate ML with neuroscience to improve understanding of the human brain and its structure, function, and/or abnormalities. Notable works so far include identifying potential biomarkers of depression from EEG signals [1], mapping cognitive function to multimodal MRI data [2, 3], and detecting brain tumors [4, 5].
Beyond task-specific applications, I am highly motivated by the development of generalized, domain-flexible models. My goal is to address the challenges of cross-domain scaling and deployability using SSL, transfer learning, and multimodal fusion in zero- or few-shot settings. At RHRI, CSU, I work on foundation models for biomedical computer vision and time-series analysis, trained on both open-access and proprietary datasets. Earlier, at BARI, Gazipur, I collected over 13,000 plant leaf images, now being used to develop a plant stress foundation model capable of generalizing across species, varieties, and stressors.