Engineer Assistant

ROHAN NARENDRA DHAMDHERE

*** ***** **., *** ********* CA-94133 +1-585-***-**** ***********@*****.***

GitHub: https://github.com/rohand24 LinkedIn: https://www.linkedin.com/in/rohandhamdhere/ EXPERIENCE

Deep Learning Engineer / Retrace Labs Inc. March 2020 – present

• Develop and deploy Image segmentation algorithms for medical X-ray image applications.

• Identifying mislabeled data in the datasets using computer vision models. Research Associate / Oak Ridge National Laboratory Dec 2018 – March 2020

• Develop and deploy ML/CV algorithms for detection and counting of buildings in satellite images.

• Counting buildings with weak supervision using only 20% of the mask labels on custom built datasets.

• Achieved a low mean absolute error of 7 for the buildings’ dataset with high building density variation.

• Model provided as tool for counting buildings on extremely large sized satellite imagery. Research Assistant / Machine Intelligence Lab at RIT Dec 2016 – Nov 2018

• Project software development related to Graph CNNs, Object detection, Image classification.

• Worked with VisualDx Inc. for improving their skin lesion identification model accuracy by 3%.

• Key projects: Graph neural networks, Traffic sign detection and Stamp feature identification. EDUCATION

Rochester Institute of Technology Aug 2016 – Oct 2018 Degree: Masters, Computer Engineering GPA: 3.90/4.0 Courses: Stanford CS231n, Deep Learning, Computer Vision, Machine Learning, Natural Language Processing. Honors: Graduate Scholarship, R.I.T. (2016-2018), Best Graduate student (Nomination) Maharashtra Institute of Technology, Pune University, India Aug 2011 – July 2015 Degree: Bachelors, Electronics and Telecommunication Engineering SKILLS

• Programming: Python, MATLAB, C++.

• Machine Learning: Scikit-Learn, Pandas, OpenCV, SVMs, Gensim, Spacy, NLTK, Weka.

• Machine Learning Frameworks: TensorFlow, Keras, Pytorch, Caffe

• Deep Learning: Object Detection, Segmentation, Classification, Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), GloVe, Graph neural networks, LSTM, GRUs PUBLICATIONS

Evolution of Graph classifiers (Thesis) (Winner- Best Paper) WNYISP 2019

• Evolving neural network architectures using custom evolutionary mutations for Graph classifiers.

• Beats architecture search time required by humans and achieves accuracy comparable to state-of-the-art models.

• Link: https://ieeexplore.ieee.org/document/8923110 General-Purpose Deep Point Cloud Feature Extractor WACV 2018

• Custom graph convolution and pooling techniques for extracting features from 3D point cloud data.

• Achieved 93% classification accuracy, best among graph methods and comparable performance to other approaches.

• Link: https://ieeexplore.ieee.org/document/8354322 Deep Learning for Philately Understanding WNYISP 2017

• Identifying stamp plate number, row/column letter for 1850s Penny Red stamps using CNNs.

• Achieved 90% accuracy for plate number identification and 98% accuracy for row/column letter identification.

• Link: https://ieeexplore.ieee.org/document/8356251 PROJECTS

Traffic Sign Detection June-July 2017

• Detection of traffic signs in custom collected Dubai based vehicle-view based images.

• Transfer learning based model using American and German traffic sign datasets.

• YOLO-v2 adaptation for detection of Traffic signs. RNN based Sentiment Analysis of Product Reviews Nov-Dec 2017

• Product category-based sentiment analysis of reviews from Amazon Review Dataset using RNNs.

• Conclusions about a product, using sentiment rating-based clusters of products to help business decisions.

• Comparison of LSTM, GRUs and TF-IDF models for classification, with TF-IDF model achieving best accuracy of 70%.

Contact this candidate