Adding empathy and emotional awareness to the next generation of conversational, multi-modal LLMs.

Course developer and facilitator for Stanford's online professional course offering of CS224W: Machine Learning with Graphs, taught by Professor Jure Leskovec. All things graphs, graph neural networks (GNNs), and PyTorch Geometric (PyG).- Led the professional adaptation, including updated lecture material, course videos, and assignments- Maintain assignment and material alignment with Stanford course + current research- Directly assist students with assignments and lecture material understanding

• Developed state-of-the art deep learning technique for identifying elephant calls in acoustic field data - leveraging curriculum learning, state-of-the-art models in audio event detection, and highly impactful data augmentation techniques. • Demonstrated promising results in support of a strong positive impact on conservation efforts• Continued research resulted in 2 papers: 1st author ICMLA-21 and 2nd author ICLMA-21• Awarded Runner Up for best 2-person project and received highest overall project score in class of over 500 (CS 231)• Selected to give a contributed talk at the 2020 ESA meeting – Harnessing the ecological data revolution

• Representation Learning with Graph Neural Networks, Generative Modeling of Graphs, and Anomaly Detection• Primary focus on increasing the expressivity of Graph Neural Networks (Identity-aware Graph Neural Networks); Accepted AAAI-21 2nd author• Secondary focus on graph anomaly detection through generative models; Application to dynamic networks, molecular generation and discovery, and social networks

Researched applications of latent variable modeling to learn low dimensional control space representations

• Designed and implemented a Wasserstein GAN to enhance the performance of a deep convolutional network trained to perform audio super-resolution (quality enhancement)• Drew inspiration from different research papers, such as GANs for image super-resolution and audio generation• Demonstrated the feasibility of applying GANs to audio-super-resolution, showing results that significantly outperform baseline methods and are comparable to past deep network models

All things ML product!At Galileo we are building the data intelligence toolchain for machine learning developers working with unstructured data -- over 80% of the world's data today.Galileo is currently powering ML teams across the Fortune 500 as well as startups across multiple industries.If you are interested in working on interesting problems at the intersection of Machine Learning and data, alongside industry and academic veterans, and at a well-funded, early stage company going after a big, real and massively underserved problem, we are currently hiring for multiple positions: - https://www.rungalileo.io/team

• Worked on large scale machine learning and deep learning to improve retrieval and ranking of Siri Search• Applied graph neural networks and knowledge graph representations to better map and understand relationships between results served to users• Analyzed and transformed petabytes of high dimensional data for machine learning models using Apache Spark

• Designed and implemented image segmentation algorithms to label the cell body and myelin sheath of nerve cells• Combined classical image segmentation algorithms with deep learning, included the UNet+ResNet architecture

• Implemented a Dynamic Programming algorithm for computing the Pareto-Frontier of tree structured networks• Greatly improved the scalability and efficiency of the Dynamic Programming algorithm: previously unsolvable networks can now be solved in seconds• Researched, designed, and implemented a divide and conquer algorithm O(NLogN) in conjunction with the DP-based approach to dramatically increase performance: entire Amazon river network from 75000sec -> 270sec• 1st author on CPAIOR-18 paper, 2nd author AAAI-18 and COMPASS-18 papers, and 3rd author of Nature Communications paper (forthcoming; currently under embargo)