Share
Director of Engineering, Tucson Technology
Build an image captioning model to generate captions of an image using CNN. Captioning the images with proper description is a popular research area of Artificial Intelligence. A good description of an image is often said as “Visualizing a picture in the mind”. The generation of descriptions from the image is a challenging task that can help and have a great impact in various applications such as usage in virtual assistants, image indexing, recommendation in editing applications, help visually impaired persons and several other natural language processing applications. In this project, we need to create a multimodal neural network which involves the concept of computer vision and Natural Language Process in recognizing the context of images and describe them in natural languages (English, etc). Deploy the model and evaluate the model on 10 different real time images
Chatbot is a conversational tool which retrieve the information from user and fullfill the user request. It was a great experience in building a conversational bot from scratch using Python and also building Amazon Alexa bot for the intents such as booking a doctor’s appointment, ordering a book, assisting in the purchase of a suitable mobile phone, finding a restaurant and many more. Firstly designed conversational flow by identifying slots, entities and worked on the configuration files. Later extracted features and classified the intents using NLP models. Finally created a database for comparing the attributes of the classified intent for achieving the desired outcome. It is not just achieving desired outcome for single intent, but also we have to shift between the intents based on user preference and maintain the dialogue flow. This project has strengthen my understanding and skills in Python, NLTK libraries, Regular Expression, Natural Language Processing (NLP), Machine Learning Algorithms and Dialogue Management.
There are many voice assistants which can translate spoken language into text, known as Automatic speech recognition. We experienced complete lifecycle in designing and implementing an E-Commerce ordering application, using speech recognition and convolutional AI techniques. It was exciting in recording our own voice samples and testing on them. Analysed the data and extracted the MFCC features of the recorded samples. Though it was a challenge in designing Convolutional and Pooling layers, but it was great learning in designing neural network architecture from scratch and train using PyTorch in selecting the product and store based on user preferences. Finally evaluated the model and saved the model to deploy in the TalentSprint server. Activated the virtual environment and imported xml file in Filezilla for uploading the deployment files and tested on web application. The final user interfaces accurately identified the product and store from the user voice commands and filtered the items based on the user requirement. It provided an excellent hands-on experience in providing end-to-end solution, starting from pre-processing the data to training and deploying the model and finally testing on the web application.
Anti Face Spoofing application is very important for protecting the sensitive data and mitigate fraud, which was implemented using AI based face and expression recognition. In this project, an EFR mobile app developed by TalentSprint, gets unlock only if both face and expression of the user gets rightly recognized. We built a deep CNN architecture using PyTorch/Keras and Pre-trained Models for facial and expression recognition and trained on large datasets. In addition to this we have used Siamese Network for identifying similarity between two images and even used the same network for facial expression. It was amazing and fun in creating our own data using EFR mobile app with different persons in our team and collecting expressions such as Happy, Sad, Surprise, Disgust, Fear, Neutral and Anger. Fine-tuned the last layers of the model and re-trained the neural network for recognizing the team’s facial and expressions. Finally evaluated the model and saved the model to deploy in the TalentSprint server. Prepared python files for the deployment and tested our face and expression in the EFR mobile app. It was an amazing experience to unlock the mobile app with our Face and Expression. We have worked on different types of architectures which enhanced our understanding on the concept of transfer learning via a pre-trained CNN to achieve a better performance.
The goal of this Mini-Hackathon was to predict whether the literacy rate was high/ medium/ low in the different districts of India. A real-world government-released dataset consisting of district-wise demographics, enrollments, and teacher indicator data was used for this project. Extensive data-preprocessing was required to clean the data, remove missing data and values, transform the data by normalization, select uncorrelated features and integrate the data into a single dataframe. After the data was pre-processed, a machine learning classifier was trained to obtain a model that could accurately predict the literacy rate (high/ medium/ low). Through this work, we were able to gain excellent experience in all aspects of data pre-processing, model building, and prediction.
The goal of this Mini-Hackathon was to classify the question based on the subcategory (like Algebra, Time and Distance, Permutations and Combinations, etc..) it belongs to. Talentsprint's aptitude questions which contain more than 20K questions data were used for this project. This project gave me an insight into using the Beautiful Soup and NLTK libraries. We have extensively used these libraries for data pre-processing. A machine learning model was trained on pre-processed data wherein given a question as input to the model it predicts the respective category of the question belongs to with >90% accuracy.
The goal of this Mini-Hackathon was to identify the unique writing styles of different authors using a machine learning-based approach. A manual, stylometric analysis using hand-crafted features was also used for comparison. This project helped me to understand the usefulness of machine learning-based text classification for automated author prediction tasks. A machine learning model was trained on textual data with corresponding author labels, from the Gutenburg database. Bag of Words and Word2Vec based approaches were used for feature representation and analysis. Text classification and author prediction were achieved with >90% accuracy.
The goal of this Mini-Hackathon is to Perform customer segmentation using bank data. The marketing department of a bank aspires to get to know its customers and suit their needs. Using bank data, create a clustering-based method to divide clients into distinct categories based on their behavioural patterns. Assist the marketing department in focusing marketing efforts on the people who are most likely to use their services and enhance their customer service.
The objective of this Mini-Hackathon was to classify images of dogs and cats, after image pre-processing and transformation. This was useful in learning image-based pre-processing techniques and achieving image classification using neural networks. A Pytorch framework was used to build and train a Convolutional Neural Network (CNN) model on a dataset of >22000 images of cats and dogs. Importantly, image classification was achieved with the CNN model, and an accuracy of >90% was obtained.
The objective of this Mini-Hackathon was to denoise the leaf images using deep learning techniques. Image denoising is useful in a variety of applications, including image restoration, image segmentation, and image classification, where recovering the original image content is critical for good results. Build an autoencoder algorithm to denoise the healthy/diseased leaf images in this project.
The goal of this Mini-Hackathon was to classify 1125 images such as sunrise, shine, rain, and cloudy using any open-source pre-trained models like Alexnet, VGG16, and Resnet50 were fine-tuned to improve the learning of the algorithm, and accuracy of >93% was obtained.
The objective of this Mini-Hackathon was to classify more than 14000 images of buildings, forests, glaciers, mountains, sea, and streets that are invariant to the transformations and can recognize the varied images. Pre-trained models like Alexnet, VGG16, and Resnet50 were fine-tuned to improve the learning of the algorithm, and accuracy of >90% was obtained.
PG Certification in AI/ML is a 9 months in-depth and comprehensive program. It enables working professionals from around the world to build AI/ML expertise from India’s top Machine Learning Lab at IIIT Hyderabad. The program is led by collaborative faculty from academia, industry and global bluechip institutions. The unique 5-step learning process of masterclass lectures, hands-on labs, mentorship, hackathons, and workshops ensure fast-track learning.