Rohin Garg Email & Phone Number

New York, NY, US

New York, New York, United States

San Diego, California, United States

Advanced Algorithms with Prof. Ramamohan PaturiML for Physical Systems with Prof. Yuanyuan ShiDeep Learning with Prof. Gary Cottrell

San Francisco, California, United States

• Designed custom loss functions to measure uncertainty based on data in the model by learning a confidence score, identified out ofdistribution data points and when to use the learned cost for AV trajectories.
• Designed and deployed models to learn a cost-field for trajectories to plan AV behavior from human driving data.
• Integrated the tensorflow model with the AV codebase in C++, to test the model on road.
• Increased efficiency of a UNET based model by 10% by changing order… Show more
• Increased efficiency of a UNET based model by 10% by changing order of up-sample and convolution layers Show less

San Diego, California, United States

Dynamic Scene Modelling for Urban Driving.

Tokyo, Japan

Designed and implemented custom data visualizations, increased search speeds in big data by about 50% by creating key-value stores for repeated searches and using better search algorithms in Splunk.

Accepted, BMVC 2021. Tackled the problem of Mode Collapse in Generative Networks, focusing on the actual cause: Catastrophic Forgetting in neural networks. We were able to improve the diversity in the generated images, in datasets like CIFAR10 and CUB200 using a GAN training method applicable to most of the existing state of the art GANs.

BEL Labs, Bengaluru, India

• Adobe Big Data Experience Lab
• Worked on Intent aware recommendations, that guide a user to a goal, i.e. a long term objective.
• Conceptualized a formal definition of Intent in the domain of user behaviour analysis.
• Introduced novel training methods, neural architectures and a custom loss function to train intent-aware agents inspired by the Reinforcement learning and time series prediction models.
• Designed evaluation functions to evaluate our work on different, more… Show more Adobe Big Data Experience Lab
• Designed evaluation functions to evaluate our work on different, more generalized datasets Show less

• REALTIME OBJECT TRACKING
• Applied real-time object tracking algorithms on IIT Kanpur surveillance cameras dataset.
• Used future frame prediction using generative adversarial networks to replace Kalman Filters in SORT algorithm.
• Improved the results using domain adaptation to IIT Kanpur’s surveillance domain.

• MUSIC INFORMATION RETRIEVAL
• Worked with Prof Vipul Arora to extract melody of different instruments from polyphonic music.
• Used the SF-NMF (Source Frequency Non-Negative Matrix Factorization) algorithm to pre-process the raw spectrogram to give enhanced inputs to Conv-RNN neural network.
• Provided a novel extension to SF-NMF, i.e. Melody specific SF-NMF that gave more flexibility to different possible melody lines to enhance by learning matrix representations.

• Edge Caching
• Implemented Virtual Networks, both sparse and dense on GENI, a testbed supported by US National Science Foundation.
• Created novel caching methods for dense small cell deployments for future 5G networks.
• Implemented Realis c web-content request pa erns using Zipf’s law for tes ng the methods using simple caching algorithms on GENI.

Master'S Degree, Computer Science And Robotics

Specialization: AI and Robotics

Bachelor'S Degree, Eecs

Activities and Societies: Robotics; Music; Tennis; Counselling Service