My recent research work mainly focuses on decentralized learning on heterogeneous data distributions. In the initial years of my PhD, I explored various research topics including Pruning, Low-Precision training, Adversarial Robustness, Spiking Neural Networks, etc.

Papers

  • Cross-feature Contrastive Loss for Decentralized Deep Learning on Heterogeneous Data
    Authors: Sai Aparna Aketi, Kaushik Roy
    Conference: IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2024)

  • Global Update Tracking: A Decentralized Learning Algorithm for Heterogeneous Data Distributions
    Authors: Sai Aparna Aketi, Abolfazl Hashemi, Kaushik Roy
    Conference: Neural Information Processing Systems (NeurIPS 2023)

  • Neighborhood Gradient Mean: An Efficient Decentralized Learning Method for Non-IID Data
    Authors: Sai Aparna Aketi, Sangamesh Kodge, Kaushik Roy
    Journal: Transactions on Machine Learning Research (TMLR 2023)

  • CoDeC: Communication-Efficient Decentralized Continual Learning
    Authors: Sakshi Choudhary, Sai Aparna Aketi, Gobinda Saha, Kaushik Roy
    Journal: Transactions on Machine Learning Research (TMLR 2024)

  • Low precision decentralized distributed training over IID and non-IID data
    Authors: Sai Aparna Aketi, Sangamesh Kodge, Kaushik Roy
    Journal: Neural Networks (2022)

  • Gradual Channel Pruning while Training using Feature Relevance Scores for Convolutional Neural Networks
    Authors: Sai Aparna Aketi, Sourjya Roy, Anand Raghunathan, Kaushik Roy
    Journal: IEEE Access, Volume 8 (2020)

  • Towards Scalable, Efficient, and Accurate Deep Spiking Neural Networks with Backward Residual Connections, Stochastic Softmax, and Hybridization
    Authors: Priyadarshini Panda, Sai Aparna Aketi, Kaushik Roy
    Journal: Frontiers in Neuroscience, Volume 14 (2020)

Workshop Papers

  • Averaging Rate Scheduler for Decentralized Learning on Heterogeneous Data
    Authors: Sai Aparna Aketi, Sakshi Choudhary, Kaushik Roy
    Conference: Tiny Paper at ICLR (2024)

  • Neighborhood Gradient Clustering: An Efficient Decentralized Learning Method for Non-IID Data
    Authors: Sai Aparna Aketi, Sangamesh Kodge, Kaushik Roy
    Conference: Federated Learning and Analytics in Practice workshop at ICML (2023)

Preprints

  • Homogenizing Non-IID datasets via In-Distribution Knowledge Distillation for Decentralized Learning
    Authors: Deepak Ravikumar, Gobinda Saha, Sai Aparna Aketi, Kaushik Roy

  • Sparse-Push: Communication- and Energy-Efficient Decentralized Distributed Learning over Directed and Time-Varying Graphs with non-IID Data
    Authors: Sai Aparna Aketi, Amandeep Singh, Jan Rabey