Nikhil Kumar J.'s Projects
In this paper, we introduce an approach to adaptive anomaly detection, which is based on a new self monitoring concept and suitable to cope with the evolving nature of the autonomous system and data.
The machine learning model can be easily fooled to incorrectly classify an input sample which was structurally and intentionally modi ed. These perturbed samples created from the original data set by making the worst case changes are called adversarial examples, and this act of fooling the model is called adversarial attacks. This vulnerability of the machine learning models to force misclassification is a major security concern since such models are deployed at various locations for various tasks.
Implemented the Alexnet neural network architecture for CIFAR10 dataset classification, using tensorflow framework.
Implemented various neural network models like Alexnet, Lenet, and VGG16 for the task of face recognition. The dataset used is a slightly different variant of the LFW dataset. Also given here is the support to save your models in h5 file format and later use it to create a tflite model to be run on embedded device.
Autoencoders are a specific type of feedforward neural networks where the input is the same as the output. They compress the input into a lower-dimensional code and then reconstruct the output from this representation. The code is a compact “summary” or “compression” of the input, also called the latent-space representation.
Presentation and Notebook for the Tech-Talk at STTP MRIIRS
Introduction to various ML and AI techniques, with hands-on projects.
Creating an end to end machine learning pipeline using Docker, Heroku, Better Code, PyTest, and more such tools.
This repo contains code for survival analysis and churn prediction of telecom customers.
Comparing the different variants of Autoencoders and evaluating their performance on the MNIST dataset.
Apr 2018 - Aug 2018 : Provides an idea of existing techniques like pruning, weight sharing and hashing to optimize the neural networks in order to achieve the optimal network which does generalization same or better than the original model.
This is a system for collaborative event detection directly on the sensor nodes. The system does not require a base station for centralized coordination or processing, and is fully trainable to recognize different classes of application-specific events. Communication overhead is reduced to a minimum by processing raw data directly on the sensor nodes and only reporting which events have been detected. an event may be anything from a malfunction of monitored machinery to an intrusion into a restricted area. The goal is to provide high-accuracy event detection at minimal energy cost in order to maximize network lifetime.
The python script scraps the number of active cases for COVID19 in India on daily basis, and sends a mail from your email to any other email address.
Fraud detection is a technique to identify unusual patterns that do not conform to the expected behaviors, and are called outliers.
The main purpose of this curated set of different data science projects is to get hands-on experience on different topics of machine learning.
Taking various NN with hidden layers from 2 to 9 and applying drop connect method on the various layers of the neural networks, to understand the effect of sparsity on the accuracy of the network. Implementation done in Tensorflow. All the networks have their pre trained weights in the respective folders. Also, the graphs showing the effect of weights being dropped over accuracy is present for all the NNs, plotted against both test and training data.
Using TensorFlow Lite, an easy solution for running machine learning models on mobile and embedded devices. It enables on‑device machine learning inference with low latency and a small binary size on Android and other embedded platforms.
This implementation is based on the Facenet paper published by Google, which proposes the idea of using inception module (basically inception network) for the task of facial recognition. This method uses a deep convolutional network trained to directly optimize the embedding itself, rather than an intermediate bottleneck layer as in previous deep learning approaches where the training is done on the complete picture rather than the face area only. To train, triplets of roughly aligned matching / non-matching face patches generated using a novel online triplet mining method. However, here we will be using the pretrained weights which is uploaded here as well for one's easy access. The benefit of this approach is much greater representational efficiency, since face recognition performance is using only 128-bytes per face.
Here, various methodologies have been discussed and tried to create a model that only includes the most important features. This has three benefits. First, the model becomes more simple to interpret. Second, we can reduce the variance of the model, and therefore overfitting. Finally, we can reduce the computational cost (and time) of training a model. The process of identifying only the most relevant features is called “feature selection.”
Complete implementation of a Variational Autoencoder in Tensorflow and understanding the behavior using MNIST dataset.
My collection of Python Programs
Inspiration is partly from the challenges of the Freiburg Hackathon 2020, as well as from our own values and desires.
Generating embeddings and finding similar images: Run inference on images to get embeddings using EfficientNet and then get K-nearest neighbors to get similar flower images.
Performed sentiment analysis and text classification on the IMDB dataset.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent