Ankit Kumar Saini Email & Phone Number

Gurugram, Haryana, India

Gurugram, Haryana, India

Gurugram, Haryana, India

The Plaksha Tech Leaders Fellowship (TLF) is a one-year, post-graduate program. TLF interweaves Artificial Intelligence (AI) and Machine Learning (ML) with real-world application, leadership, and mentorship.TLF is built on four intertwined threads: Data-X coursework (AI and ML, design and systems thinking), Real-world Projects (Challenge Lab and Capstone Project), Guest Sessions by domain experts, business leaders and entrepreneurs, and Leadership and Mentorship.

West Lafayette, Indiana, United States

In my stint as a research scholar, I have extensively worked on training and evaluating the performance of convolutional based deep learning models for various agricultural applications such as the classification and detection of crop weeds and diseases. 1. Early Stage Identification of Weeds in Crop Fields:- Fine-tuned 3 popular deep learning models, InceptionV3, ResNet50, and EfficientNetB4, using pre-trained ImageNet weights via Transfer Learning to classify images into four weed classes: Ragweed, Pigweed, Cocklebur, and Foxtail. Each model was fine-tuned separately in TensorFlow and PyTorch for comparison.- EfficientNetB4 achieved the highest F1-score of 98%, followed by ResNet50 (97%), and InceptionV3 (95%).- Performed error analysis using GradCAM and Saliency Maps to diagnose prediction errors.2. Generalization Performance of Deep Learning Models Trained on Images With Uniform Backgrounds:- Used the PlantVillage dataset, consisting of plant disease images with uniform backgrounds, split across 38 categories.- Generated eight new datasets with different backgrounds (Soil, Field, Grass, Straw, Red, Blue, White, and Black) using OpenCV, resulting in nine versions of the PlantVillage dataset.- Trained 4 popular deep learning models (MobileNetV2, InceptionV3, ResNet50, and EfficientNetB0) using Transfer Learning. Evaluated the generalization performance of each model by training on one dataset and testing on the remaining datasets.- EfficientNetB0 demonstrated superior performance with the highest test set accuracy of 92%. The model's accuracy dropped by at least 2% when evaluated on other datasets. Varying accuracy (90% to 85%) were observed on other backgrounds.- Results indicate deep learning models should not be trained on images with uniform backgrounds for field applications.3. Cassava Leaf Disease Classification:- Trained EfficientNetB4 on 25,000 images to classify leaf disease into 5 classes.- Deployed the model as a web app using Flask.

Mahesana, Gujarat, India

Electrical Engineer - Production Engineering Department (Weld)I worked on a greenfield project, initially joining as a Graduate Engineer Trainee (GET).1. English to Hindi Text Translation For Weld PE Team Operators:- Developed a deep learning model to translate English text into Hindi for weld PE team operators.- Built a Seq2seq model with an LSTM-based encoder and decoder using TensorFlow in Python.- Trained the network end-to-end using the categorical cross-entropy loss function and the Adam optimizer. EarlyStopping callback was used to prevent overfitting on the training dataset.2. Hole Detection in Sheet Metal (Cowl Side Panel):- Trained a ResNet50 model to assist the operators in detecting the presence of a hole in the cowl side panel of different automobile models.- Addressed the issue of operators frequently selecting the wrong part due to the similar appearance of cowl side panels for Swift and Baleno models, leading to manual rework and scrapping of the car frame.- The distinguishing feature was an additional hole in the Baleno panel. The ResNet50 model, trained in TensorFlow, effectively detected this hole, easing the identification process.3. Other primary responsibilites were- Set up data collection systems for each line/station- Analyzed data collected from different lines/stations- Installed new equipment and fine-tuned parameters- Industrial robot teaching for spot welding- Conducted new vehicle model trials- Planned and procured electrical spares- Conducted trials and testing of weld line equipment

Certificate Of Completion, Data Scientist Nanodegree

- Learn the data science process, including how to build effective data visualizations and communicate with various stakeholders..

Bachelor Of Technology - Btech, Electrical Engineering, 8.47

Education record