publications

2025

1. IJCNN

   

   DeBUGCN – Detecting Backdoors in CNNs Using Graph Convolutional Networks

   Akash Vartak, Khondoker Murad Hossain, and Tim Oates

   2025

   Abs Bib HTML PDF

   

   Deep neural networks (DNNs) are becoming commonplace in critical applications, making their susceptibility to backdoor (trojan) attacks a significant problem. In this paper, we introduce a novel backdoor attack detection pipeline, detecting attacked models using graph convolution networks (DeBUGCN). To the best of our knowledge, ours is the first use of GCNs for trojan detection. We use the static weights of a DNN to create a graph structure of its layers. A GCN is then used as a binary classifier on these graphs, yielding a trojan or clean determination for the DNN. To demonstrate the efficacy of our pipeline, we train hundreds of clean and trojaned CNN models on the MNIST handwritten digits and CIFAR-10 image datasets, and show the DNN classification results using DeBUGCN. For a true In-the-Wild use case, our pipeline is evaluated on the TrojAI dataset which consists of various CNN architectures, thus showing the robustness and model-agnostic behaviour of DeBUGCN. Furthermore, on comparing our results on several datasets with state-of-the-art trojan detection algorithms, DeBUGCN is faster and more accurate.

   @article{vartak2025debugijcnn,
     title = {DeBUGCN -- Detecting Backdoors in CNNs Using Graph Convolutional Networks},
     author = {Vartak, Akash and Hossain, Khondoker Murad and Oates, Tim},
     year = {2025},
     conference = {International Joint Conference on Neural Networks},
     url = {https://arxiv.org/abs/2502.18592},
     language = {English},
   }

2. arXiv

   

   DeBUGCN – Detecting Backdoors in CNNs Using Graph Convolutional Networks

   Akash Vartak, Khondoker Murad Hossain, and Tim Oates

   2025

   Abs Bib HTML PDF

   

   Deep neural networks (DNNs) are becoming commonplace in critical applications, making their susceptibility to backdoor (trojan) attacks a significant problem. In this paper, we introduce a novel backdoor attack detection pipeline, detecting attacked models using graph convolution networks (DeBUGCN). To the best of our knowledge, ours is the first use of GCNs for trojan detection. We use the static weights of a DNN to create a graph structure of its layers. A GCN is then used as a binary classifier on these graphs, yielding a trojan or clean determination for the DNN. To demonstrate the efficacy of our pipeline, we train hundreds of clean and trojaned CNN models on the MNIST handwritten digits and CIFAR-10 image datasets, and show the DNN classification results using DeBUGCN. For a true In-the-Wild use case, our pipeline is evaluated on the TrojAI dataset which consists of various CNN architectures, thus showing the robustness and model-agnostic behaviour of DeBUGCN. Furthermore, on comparing our results on several datasets with state-of-the-art trojan detection algorithms, DeBUGCN is faster and more accurate.

   @article{vartak2025debugcndetectingbackdoors,
     title = {DeBUGCN -- Detecting Backdoors in CNNs Using Graph Convolutional Networks},
     author = {Vartak, Akash and Hossain, Khondoker Murad and Oates, Tim},
     year = {2025},
     eprint = {2502.18592},
     archiveprefix = {arXiv},
     primaryclass = {cs.CV},
     url = {https://arxiv.org/abs/2502.18592},
     language = {English},
   }

2022

1. Masters Thesis

   

   Using Text to Improve Classification of Man-Made Objects

   Akash A. Vartak

   2022

   Copyright - Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual underlying works; Last updated - 2023-08-04

   Abs Bib HTML PDF

   

   People identify man-made objects by their visual appearance and the text on them e.g., does a bottle say water or shampoo? We use text as an important visual cue to help distinguish between similar looking objects. This thesis explores a novel joint model of visual appearance and textual cues for image classification.We perform this in three functions - (a) Isolating an object in an input image; (b) Extracting text from the image; (c) Training a joint vision/text model. We simplify the task by extracting text separately and presenting it to the model in machine readable format. Such a joint model has utility in many real world challenges where language is interpreted through a sensory perception like vision or sound.The aim of the research is to understand whether visual percepts, when understood in the context of extracted language, will provide a better classification of image objects than using only pure vision to perform image classification. In conclusion, we show that joint classifier models can successfully make use of text present in images to classify objects, provided that the extracted text from images is of high quality and we have the number of images proportional to the number of classification classes.

   @phdthesis{vartak2022thesis,
     author = {Vartak, Akash A.},
     year = {2022},
     title = {Using Text to Improve Classification of Man-Made Objects},
     journal = {ProQuest Dissertations and Theses},
     pages = {55},
     note = {Copyright - Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual underlying works; Last updated - 2023-08-04},
     keywords = {Classification; Computer vision; Deep learning; Machine learning; Natural language processing; Text extraction; Computer science; 0984:Computer science},
     isbn = {9798837525858},
     language = {English},
     url = {http://proxy-bc.researchport.umd.edu/login?url=https://www.proquest.com/dissertations-theses/using-text-improve-classification-man-made/docview/2695026845/se-2},
   }

2017

1. IJCS

   

   A Survey on Promotional and Base Level Forecasting using ARIMA

   Aakash Raina, Akash Vartak, Pravin R Patil, Yash Katariya, and Yash Lahoti

   Jan 2017

   Abs Bib

   

   Forecasting is the process of making predictions of the future based on past and present data and most commonly by analysis of trends. Sales forecasting is the process of estimating future sales. To make informed business decisions and predict short-term and long-term performance an accurate sales forecast is a must. Promotions and targeted marketing is used by many businesses to increase the demand for or visibility of their product. Such promotions normally require increased expenses, loss of revenue and maybe additional costs. But having incurred all the financial and nonfinancial losses business owners need to determine the value and benefit of these promotions. So one way to evaluate promotions is to analyze the historical data using time series analysis techniques. This paper briefly explores time series analysis for promotional and base level forecasting.

   @misc{vartak2017ArimaSurvey,
     title = {A Survey on Promotional and Base Level Forecasting using ARIMA},
     author = {Raina, Aakash and Vartak, Akash and Patil, Pravin R and Katariya, Yash and Lahoti, Yash},
     year = {2017},
     month = jan,
     keywords = {Promotional analysis; Demand Analysis; Forecasting; Sales; Time series analysis; ARIMA},
     journal = {International Journal of Computer Systems},
     isbn = {2394-1065},
     language = {English},
     url = {https://akash-vartak.github.io/A_Survey_on_Promotional_and_Base_Level_Forecasting_using_ARIMA.pdf},
   }