Projects
- Inserting Stealthy Syntactic Backdoors which are not perceivable by the Humans during Input checks and audits
- Using T5 Model along with SCPN to generate paraphrased sentences with minor syntactic changes so as to trigger backdoor only when inputs with these given syntactic rule are passed.
- Resulted in increased ASR (Attack success rate) to 96.73% for given stealthy syntactical attacks
View here
- Created various iterations of multimodal neural network to identify hateful intent from the large corpus of memes offered as a part of NeurIPS 2020 Challenge (Proposed by Facebook AI)
View here
- Built and evaluated a recommender system on MovieLens large dataset using PySpark’s ALS module.
- Proposed a baseline popularity model for cold start strategy.
- Extended the system by implementing and evaluating on LightFM and contrasting the performance with ALS model.
- Researched and trained a self-supervised learning model for depth estimation on KITTI dataset by experimenting with various iterations of UNet Architecture involving Encoders (ResNet, ConvNext ) and Up-sampling decoders like ESPCN achieving a rms of 3.7
- Enhanced our model using Mask R-CNN pretrained on COCO Dataset to get semantic map which helped us reduce artifact problem in the disparity map and consequently reducing the overall loss for the model to 3.485
View here
- Formulated a compensation model using regularized regression on Salary data from Stack overflow (100k, 300 Column dataset).
- Experimented with various Machine Learning techniques which included PCA, KMeans Clustering and Classification (Random Forest, Support Vector Classifier, Logistic Regression) on a stack overflow survey data
- Authored an Open Sourced NPM Library for asynchronous batch operations which provided a 17% improvement in the execution time and better handling of error catching and response compared to the native Promise.all Method
