Darshan Patil

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Copy DOI

10.2139/ssrn.3568717

Cite

Citations (2)

Disentangling 3D Prototypical Networks For Few-Shot Concept Learning

Mihir Prabhudesai Shamit Lal Darshan Patil Hsiao-Yu Fish Tung Adam W. Harley

We present neural architectures that disentangle RGB-D images into objects' shapes and styles and a map of the background scene, and explore their applications for few-shot 3D object detection and few-shot concept classification. Our networks incorporate architectural biases that reflect the image formation process, 3D geometry of the world scene, and shape-style interplay. They are trained end-to-end self-supervised by predicting views in static scenes, alongside a small number of 3D object boxes. Objects and scenes are represented in terms of 3D feature grids in the bottleneck of the network. We show that the proposed 3D neural representations are compositional: they can generate novel 3D scene feature maps by mixing object shapes and styles, resizing and adding the resulting object 3D feature maps over background scene feature maps. We show that classifiers for object categories, color, materials, and spatial relationships trained over the disentangled 3D feature sub-spaces generalize better with dramatically fewer examples than the current state-of-the-art, and enable a visual question answering system that uses them as its modules to generalize one-shot to novel objects in the scene.

Feature (linguistics)

Cite

Citations (0)

An Overview of Text Representation Techniques in Text Classification using Deep Learning Models

2022 3rd International Conference for Emerging Technology (INCET) (2022)

Darshan Patil Reena Lokare Sunita Patil

Organizations handling huge amount of data needs to preserve privacy of the documents. Every customer has the rights to ask for privacy of their documents. These documents can be classified into different categories like private, public, confidential, etc. Using suitable text classification methods, documents can be classified into different categories. Various approaches like machine learning (ML), deep learning (DL), natural language processing (NLP) is available. Machine learning algorithms are showing acceptable performance but doesn't work well when data grows in size. In this paper, convolutional neural network(CNN) which is a deep learning model, is used to classify documents into different categories. Deep learning models are beneficial over machine learning models in terms of performance and volume of data to be classified. The performance of the model is evaluated and obtained acceptable performance.

Representation

10.1109/incet54531.2022.9825389

Cite

Citations (2)

Towards modular and programmable architecture search

arXiv (Cornell University) (2019)

Renato Negrinho Darshan Patil Nghia Le Daniel C. Ferreira Matthew R. Gormley

Neural architecture search methods are able to find high performance deep learning architectures with minimal effort from an expert. However, current systems focus on specific use-cases (e.g. convolutional image classifiers and recurrent language models), making them unsuitable for general use-cases that an expert might wish to write. Hyperparameter optimization systems are general-purpose but lack the constructs needed for easy application to architecture search. In this work, we propose a formal language for encoding search spaces over general computational graphs. The language constructs allow us to write modular, composable, and reusable search space encodings and to reason about search space design. We use our language to encode search spaces from the architecture search literature. The language allows us to decouple the implementations of the search space and the search algorithm, allowing us to expose search spaces to search algorithms through a consistent interface. Our experiments show the ease with which we can experiment with different combinations of search spaces and search algorithms without having to implement each combination from scratch. We release an implementation of our language with this paper.

ENCODE

Interface (matter)

Implementation

Source

Cite

Citations (6)

Transparent precision: Explainable AI empowered breast cancer recommendations for personalized treatment

IAES International Journal of Artificial Intelligence (2024)

Reena Lokare Jyoti Wadmare Sunita Patil Ganesh Wadmare Darshan Patil

<p>Breast cancer stands as a prevalent global concern, prompting extensive research into its origins and personalized treatment through Artificial Intelligence (AI)-driven precision medicine. However, AI's black box nature hinders result acceptance. This study delves into Explainable AI (XAI) integration for breast cancer precision medicine recommendations. Transparent AI models, fuelled by patient data, enable personalized treatment recommendations. Techniques like feature analysis and decision trees enhance transparency, fostering trust between medical practitioners and patients. This harmonizes AI's potential with the imperative for clear medical decisions, propelling breast cancer care within the precision medicine era. This research work is dedicated to leveraging clinical and genomic data from samples of metastatic breast cancer. The primary aim is to develop a machine learning (ML) model capable of predicting optimal treatment approaches, including but not limited to hormonal therapy, chemotherapy, and anti-HER2 therapy. The objective is to enhance treatment selection by harnessing advanced computational techniques and comprehensive data analysis. A decision tree model developed here for the prediction of suitable personalized treatment for breast cancer patients achieves 99.87% overall prediction accuracy. Thus, the use of XAI in healthcare will build trust in doctors as well as patients.</p>

10.11591/ijai.v13.i3.pp2694-2702

Cite

Citations (1)

ELECTRICITY THEFT DETECTIONUSING MACHINE LEARNING: A REVIEW

International Research Journal of Modernization in Engineering Technology and Science (2023)

Rutuja Avhad Vaishnavi Biradar Darshan Patil

Advanced metering structure is one of the important factors of smart grid and offers an essential link between consumers and their loads, grid, and rubricstion and storehouse coffers.Electricity theft, one of the crucial concern in AMI, causes million bone profit loss every time in developing and developed countries.In this paper, star element Analysis( PCA) grounded electricity theft discovery scheme is proposed.PCA is used to transfigure a high dimensional dataset into a low dimensional dataset.Using top factors, anomaly score is calculated and compared with a predefined threshold value.The proposed scheme is tested under different attack script using real dataset.The results show that the proposed scheme detects electricity theft attacks with high discovery rate.One of the main contributors to the nontechnical power losses is the loss due to electricity theft.In developing countries like Pakistan, the fiscal losses due to electricity theft are veritably high and pose a veritably serious trouble to the country's profitable stability.Electricity theft by cyber attacks to hurt grid measures in advanced metering structure affects fiscal losses to serviceability every time.In this paper we're using metering data grounded Extreme Gradient Boosting.There are numerous operations available for electricity theft discovery.There are many machine literacy ways are discovered for electricity theft discovery.This study, thus, com-pares the prophetic delicacy of several machine literacy styles including Logistic Retrogression( LR), K-Nearest Neighbor Algorithm,( K-NN), Support Vector Machines( SVM), and Neural Networkshop( NNet) for prognosticating the electricity thefts in a concrete model.

10.56726/irjmets36356

Cite

Citations (0)

Privacy Preserving Document Classification using Convolution Neural Network- A Deep Learning Approach

Darshan Patil Reena Lokare Sunita Patil

Increased digitization in nearly every sector demands huge data storage requirements. Every person upload tons of information related to themselves on Internet through some mobile or web application, knowingly or sometimes unknowingly. Such increasing personal data storage requirement has created data privacy issues. There is no law which prohibits someone from using personal information of an individual. India is still in the process of preparing personal data protection law, whereas European Union's data protection regulation has already took place in the year 2018. Some organizations are in the process of developing applications which can check whether a document is personal or non-personal. Such applications can be developed with the help of deep learning models such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Term Short Memory (LSTM), etc. This research focuses on different text representation techniques required to represent text in text classification problems such as private data classification, sentiment analysis, language detection, online abuse detection, recommendations systems, to name a few. Having represented text in different formats, helps in increasing accuracy of classification algorithms.

Upload

10.1109/icast55766.2022.10039513

Cite

Citations (0)

Disentangling 3D Prototypical Networks For Few-Shot Concept Learning

arXiv (Cornell University) (2020)

Mihir Prabhudesai Shamit Lal Darshan Patil Hsiao-Yu Fish Tung Adam W. Harley

We present neural architectures that disentangle RGB-D images into objects' shapes and styles and a map of the background scene, and explore their applications for few-shot 3D object detection and few-shot concept classification. Our networks incorporate architectural biases that reflect the image formation process, 3D geometry of the world scene, and shape-style interplay. They are trained end-to-end self-supervised by predicting views in static scenes, alongside a small number of 3D object boxes. Objects and scenes are represented in terms of 3D feature grids in the bottleneck of the network. We show that the proposed 3D neural representations are compositional: they can generate novel 3D scene feature maps by mixing object shapes and styles, resizing and adding the resulting object 3D feature maps over background scene feature maps. We show that classifiers for object categories, color, materials, and spatial relationships trained over the disentangled 3D feature sub-spaces generalize better with dramatically fewer examples than the current state-of-the-art, and enable a visual question answering system that uses them as its modules to generalize one-shot to novel objects in the scene.

Feature (linguistics)

10.48550/arxiv.2011.03367

Cite

Citations (8)

Selective Attention UNet for Segmenting Liver Tumors

2022 International Conference on Cyber-Physical Social Intelligence (ICCSI) (2023)

Darshan Patil Gopal Sakarkar Pearl Khatri Atharva Khedkar Hong‐Seng Gan

In order to segregate liver tumours in medical imaging applications, a novel architecture called Selective Attention UNet is proposed in this article. The suggested architecture, which is based on the well-known UNet architecture, has a selective attention module that enables the network to concentrate on crucial tasks while suppressing unnecessary ones. Link skipping between the encoder and decoder routes is another element of the design that enables the network to effectively segment data using both low-level and high-level attributes. On the publicly accessible LiTS dataset, we assessed the performance of the suggested architecture and contrasted it with four fundamental models: FCN, UNet, UNet++, and SegNet. The Dice Similarity Coefficient (DSC) of 0.89 a mean IOU of 0.76 obtained in our experiments demonstrates that the suggested architecture beats all baseline models in terms of accuracy and robustness criteria. The project is accessible at: https://github.com/darshan8850/Liver-tumor-Segmentation

Robustness

Dice

10.1109/iccsi58851.2023.10303778

Cite

Citations (0)

Towards modular and programmable architecture search

arXiv (Cornell University) (2019)

Renato Negrinho Matthew R. Gormley Geoffrey J. Gordon Darshan Patil Nghia Le

Neural architecture search methods are able to find high performance deep learning architectures with minimal effort from an expert. However, current systems focus on specific use-cases (e.g. convolutional image classifiers and recurrent language models), making them unsuitable for general use-cases that an expert might wish to write. Hyperparameter optimization systems are general-purpose but lack the constructs needed for easy application to architecture search. In this work, we propose a formal language for encoding search spaces over general computational graphs. The language constructs allow us to write modular, composable, and reusable search space encodings and to reason about search space design. We use our language to encode search spaces from the architecture search literature. The language allows us to decouple the implementations of the search space and the search algorithm, allowing us to expose search spaces to search algorithms through a consistent interface. Our experiments show the ease with which we can experiment with different combinations of search spaces and search algorithms without having to implement each combination from scratch. We release an implementation of our language with this paper.

ENCODE

Interface (matter)

Combinatorial search

Source

Cite

Citations (11)

Papers

Citation

H-Index

i-10 index

Papers

Citation

H-Index

i-10 index