Piyush Tiwary

selected publications

1. 

   Few Shot Generative Domain Adaptation Via Inference-Stage Latent Learning in GANs

   Arnab Kumar Mondal, Piyush Tiwary, Parag Singla, and 1 more author

   In International Conference on Learning Representation (ICLR), 2023

   Abs Bib HTML

   In this work, our objective is to adapt a Deep generative model trained on a large-scale source dataset to multiple target domains with scarce data. Specifically, we focus on adapting a pre-trained Generative Adversarial Network (GAN) to a target domain without re-training the generator. Our method draws the motivation from the fact that out-of-distribution samples can be ‘embedded’ onto the latent space of a pre-trained source-GAN. We propose to train a small latent-generation network during the inference stage, each time a batch of target samples is to be generated. These target latent codes are fed to the source-generator to obtain novel target samples. Despite using the same small set of target samples and the source generator, multiple independent training episodes of the latent-generation network results in the diversity of the generated target samples. Our method, albeit simple, can be used to generate data from multiple target distributions using a generator trained on a single source distribution. We demonstrate the efficacy of our surprisingly simple method in generating multiple target datasets with only a single source generator and a few target samples.

   @inproceedings{mondal2023fewshoticlr,
     title = {Few Shot Generative Domain Adaptation Via Inference-Stage Latent Learning in GANs},
     author = {Mondal, Arnab Kumar and Tiwary, Piyush and Singla, Parag and AP, Prathosh},
     booktitle = {International Conference on Learning Representation (ICLR)},
     year = {2023},
     url = {https://openreview.net/forum?id=sCYXJr3QJM8},
   }

2. 

   Minority Oversampling for Imbalanced Data via Class-Preserving Regularized Auto-Encoders

   Arnab Kumar Mondal, Lakshya Singhal, Piyush Tiwary, and 2 more authors

   In International Conference on Artificial Intelligence and Statistics (AISTATS), 2023

   Abs Bib HTML

   Class imbalance is a common phenomenon in multiple application domains such as healthcare, where the sample occurrence of one or few class categories is more prevalent in the dataset than the rest. This work addresses the class-imbalance issue by proposing an over-sampling method for the minority classes in the latent space of a Regularized Auto-Encoder (RAE). Specifically, we construct a latent space by maximizing the conditional data likelihood using an Encoder-Decoder structure, such that oversampling through convex combinations of latent samples preserves the class identity. A jointly-trained linear classifier that separates convexly coupled latent vectors from different classes is used to impose this property on the AE’s latent space. Further, the aforesaid linear classifier is used for final classification without retraining. We theoretically show that our method can achieve a low variance risk estimate compared to naive oversampling methods and is robust to overfitting. We conduct several experiments on benchmark datasets and show that our method outperforms the existing oversampling techniques for handling class imbalance.

   @inproceedings{mondal2023OSA,
     author = {Mondal, Arnab Kumar and Singhal, Lakshya and Tiwary, Piyush and Singla, Parag and Prathosh, AP},
     booktitle = {International Conference on Artificial Intelligence and Statistics (AISTATS)},
     year = {2023},
   }

3. 

   FadeLoc: Smart Device Localization for Generalized κ- μ Faded IoT Environment

   Ankur Pandey*, Piyush Tiwary*, Sudhir Kumar, and 1 more author

   IEEE Transactions on Signal Processing, 2022

   Abs Bib HTML

   In this paper, we propose FadeLoc a novel method for localizing smart devices in an Internet of Things (IoT) environment, based on the Received Signal Strength (RSS), and a generic κ- μ fading model where κand μdenote the fading parameters. The RSS-based localization is challenging because of noise, fading, and non-line-of-sight (NLOS) effects, thus necessitating an appropriate fading model to best fit the varying RSS values. The advantage of a generic fading model is that it can accommodate all existing fading distributions based on the estimate of κand μ. Hence, the localization can be performed for any fading environment. We derive the maximum likelihood estimate of the smart device location using a generic κ- μ fading model considering the Large and Small approximations of modified first-order Bessel function and propose an adaptive order selection method with high localization accuracy and faster convergence. We also analyze the convergence of the gradient ascent method for the κ- μ fading model. The proposed method is evaluated on a simulated κ- μ fading environment, real outdoor environment, and a complex indoor fading environment. The average localization errors are 2.07 m, 3.5 m, and 0.5 m, respectively, for the three experimental settings, outperforming the state-of-the-art localization methods in the presence of fading.

   @article{9797867,
     author = {Pandey*, Ankur and Tiwary*, Piyush and Kumar, Sudhir and Das, Sajal K.},
     journal = {IEEE Transactions on Signal Processing},
     title = {FadeLoc: Smart Device Localization for Generalized ${\kappa {-} \mu}$ Faded IoT Environment},
     year = {2022},
     volume = {70},
     pages = {3206-3220},
     keywords = {},
     doi = {10.1109/TSP.2022.3183527},
     url = {https://ieeexplore.ieee.org/document/9797867},
     issn = {1941-0476},
   }