Aditya Dutt, Ph.D.
Postdoctoral Associate, Electrical & Computer Engineering, University of Florida
GatorSense · Machine Learning and Sensing Lab
I develop machine learning for scientific data where sensors are heterogeneous and labels are scarce: aligning modalities in shared latent spaces, generating realistic synthetic data to train downstream models, and structuring embeddings so they transfer across domains. Applications include remote sensing, plant root phenotyping from X-ray CT, crop phenotyping from drone imagery, computational pathology, and socioeconomic mapping.
I am a Postdoctoral Associate in the Machine Learning and Sensing Lab (GatorSense) at the University of Florida, working with Dr. Alina Zare. I received my Ph.D. in Computer and Information Science and Engineering from UF in 2025.
On the job market: tenure-track faculty positions, 2026–27.
Research
(JSTARS 2022)
Multimodal alignment and fusion
Mapping hyperspectral, LiDAR, SAR and optical data into one class-discriminative space, so one model serves every sensor and a missing sensor can be predicted from the others.
(CANVAS 2026)
Generative modeling for scientific data
Translating simulated measurements into realistic ones: sampled root skeletons condition generated 3D root geometry, which is then rendered as synthetic X-ray CT of soil cores, so segmentation models can be trained without costly manual annotation.
(arXiv 2026)
Structured latent spaces under limited supervision
Contrastive and triplet objectives that shape embedding geometry, autoencoder representations for downscaling survey indicators across India, and weakly supervised graph networks trained on synthetic and incomplete labels.
Dissertation
A Multimodal Alignment Network for Domain Translation and Fusion
Ph.D., Computer and Information Science and Engineering, University of Florida, May 2025 · Advisor: Dr. Alina Zare
Introduces COMMANet, a contrastive multimodal alignment network that uses triplet networks to map heterogeneous sensors (hyperspectral, LiDAR, SAR, optical) into one shared latent space. A single model performs both classification and domain translation, and the shared space generates aligned multimodal synthetic data for low-resource settings.
Publications
- 2026Joint utilization of geospatial and census proxies for autoencoder-assisted downscaling (JUGAAD) of socioeconomic indicators in India. A. Dutt, P. Gader, A. Singh. arXiv preprint. preprint arXiv
- 2025Multi-task learning with multi-annotation triplet loss for improved object detection. M. Zhou, A. Dutt, A. Zare. IEEE International Geoscience and Remote Sensing Symposium (IGARSS). arXiv
- 2025A multimodal alignment network for domain translation and fusion. A. Dutt. Ph.D. Dissertation, University of Florida.
- 2023Wavelet multiresolution analysis based speech emotion recognition system using 1D CNN LSTM networks. A. Dutt, P. Gader. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 31, 2043–2054. IEEE Xplore
- 2022
- 2018Colour image segmentation with histogram and homogeneity histogram difference using evolutionary algorithms. S. Kumar, M. Pant, M. Kumar, A. Dutt. International Journal of Machine Learning and Cybernetics, 9, 163–183.
Presentations
- 2026FUSION 1.0: Integrating kidney whole slide histology with 10x Visium spatial transcriptomics for FTU-level analysis. I. Ansari, A. Dutt, P. Sarder. Poster, UF College of Medicine Celebration of Research.
- 2025Synthetic root architecture modeling–guided segmentation of real X-ray CT point clouds of soil cores using weak ground truth labels in switchgrass. J. Diaz, A. Dutt, A. Zare, et al. Oral, CANVAS 2025, Utah.
- 2025Getting to the core of roots: segmentation of CT-scan point clouds using weak ground truths. J. Diaz, A. Dutt, A. Zare. Poster, AI Days 2025, UF.
- 2024COMMANet: a contrastive multi-modal alignment network for domain translation and fusion. A. Dutt. Poster, AI Days 2024, UF.
Experience
Diffusion and graph-based generative models for X-ray CT root segmentation (DSFAS); spatiotemporal trait prediction for switchgrass from drone imagery.
Project manager for computational pathology software integrating kidney histology with spatial transcriptomics (FUSION 1.0).
Multi-sensor translation and fusion with missing data; root segmentation from CT; socioeconomic downscaling for rural India.