Publications

Comparing Adversarial Unsupervised Domain Adaptation to Zero-Shot Classification in Contrastive Language-Image Pre-training Embedding Space
Kaustubh Deshpande, Advisor: Ying Nian Wu
View Publication

Enlisting 3D Crop Models and GANs for More Data Efficient and Generalizable Fruit Detection
Acknowledgement for synthetic image generation following feature extraction on a smaller real image dataset
View Publication

Accurate Detection of RNA Stem-Loops in Structurome Data Reveals Widespread Association with Protein Binding Sites
Pierce Radecki, Rahul Uppuluri, Kaustubh Deshpande, Sharon Aviran
https://doi.org/10.1101/2021.04.28.441809
View Publication

Microfluidic cap-to-dispense (μCD): a universal microfluidic–robotic interface for automated pipette-free high-precision liquid handling
Wang Jingjing, Ka Deng, Chuqing Zhou, Zecong Fang, Conary Meyer, Kaustubh Deshpande, Zhihao Li, Xianqiang Mi, Qian Luo, Bruce D. Hammock, Cheemeng Tan, Yan Chen, and Tingrui Pan. “Microfluidic Cap-to-dispense (μCD): A Universal Microfluidic–robotic Interface for Automated Pipette-free High-precision Liquid Handling.” Lab on a Chip, 19(20), 3405–3415. https://doi.org/10.1039/c9lc00622b
View Publication

Research Experience

• Tixr :
  • Current research focuses on making LLMs more truthful by mitigating hallucinations, robust to adversarial attacks, and aligning these models with human values. The end goal is to utilize resulting models to fulfill client tasks such as financial report generation, chatbot assistance for navigating our product.
  • Gained hands-on experience cutting open transformer architectures, implementing RLHF, RLAIF, DPO, and fine-tuning models for deployment and inference in the cloud, particularly on AWS.
  • A bigger picture research interest of mine is efficient pre training techniques and knowledge distillation. Result would be cheaper model development and smaller model size making this technology much more accessible to people around the world.
• Amazon :
  • Worked with time series transformers, time series classification and regression analysis to improve existing forecast model
  • Wrote an internal research paper
• Plant AI & Biophysics Lab :
  • At the PAIBL my research was centered around generating and utilizing synthetic data for computer vision tasks.
  • Working on various object detection and instance segmentation tasks, I gained familiarity implementing and customizing the Mask R-CNN architecture.
• Computational RNA Genomics Lab
  • My research at this lab primarily involved statistical analysis of large RNA data sets and software development of the lab’s primary product: PatteRNA.
  • I worked on the experimentation and integration of a binary classifier into the PatteRNA pipeline which resulted in a rapid and accurate method for automatically detecting families of RNA structure motifs.
• Micro-nano Innovations Lab
  • At the MiNi lab, the aim of our project was to develop a robotic–microfluidic interface for complete pipette-free liquid handling automation.
  • I worked on the computer vision segment of this project and the software I developed was utilized by a DOBOT robotic arm to identify chemicals in a laboratory setting and dispense them as needed.

Back to home page ⮕