Nick Kipshidze
AI/ML Engineer
+
Full Stack Dev
Simple blob classification using PyTorch.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset. The dataset was generated by one of SciKit-learn's load dataset functions.
Classifying circular data points with PyTorch.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset. The dataset was generated by one of SciKit-learn's load dataset functions.
Classifying spiral data points with PyTorch.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset. The dataset was generated by one of SciKit-learn's load dataset functions.
Classifying spiral data points with PyTorch.
With this project I attempted overfitting the datapoints with my complicated neural network, over longer period of time, with more epochs/iterations.
Classifying spiral-shaped data points with PyTorch.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset.
Classifying gaussian quantiles dataset with PyTorch.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset.
Model fitted to Sklearn's "Iris" toy dataset.
Due to the simplicity of the dataset, the model can predict labels for the testing features at 100% accuracy. By graphic the confusion matrix, we can get a perfect graph.
Model fitted to Sklearn's "Wine" toy dataset.
Classify wine based on it's features such as hue, saturation, color and so on. The model works almost perfectly, predicting the true label at a near perfect accuracy.
Model fitted to Sklearn's digits toy dataset.
Recognise handwritten numbers based on 8x8 resolution image. The model works almost perfectly, predicting the true label at a near perfect accuracy.
Model fitted to Sklearn's "cancer" toy dataset.
Recognize cancer given numerical attributes. The model works almost perfectly, with the accuracy of 96.00% on the testing dataset.
Computer vision model that can classify (recognize) handwritten digits. I used the MNIST dataset which consists of 60,000 training images and 10,000 testing images.
The model works at 93.75% accuracy, and it is very lightweight.
Computer vision model that can classify (recognize) handwritten digits. I used the MNIST dataset which consists of 60,000 training images and 10,000 testing images.
The model works at 93.75% accuracy, and it is not very lightweight.
Computer vision model that can classify (recognize) different clothing. I used the fashion MNIST dataset and the TinyVGG convolutional neural network architecture as a reference.
The model works at 88.43% accuracy, considering the similarity of different clothing classes, the accuracy is reasonable.
Simple blob classification using TensorFlow.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset. The dataset was generated by one of SciKit-learn's load dataset functions.
Classifying circular data points with TensorFlow.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset. The dataset was generated by one of SciKit-learn's load dataset functions.
Classifying spiral data points with TensorFlow.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset.
Classifying spiral shaped data points using TensorFlow.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset.
Classifying gaussian quantiles dataaset using TensorFlow.
The model can classify a datapoint's color based on it's coordinates, which are the features in the dataset.
Project was built using Python and PyTorch as the primary framework. The dataset used is the MNIST handwritten digits dataset, the model was trained for 10 epochs using the Adam optimizer and categorical cross-entropy loss function. The convolutional neural network architecture used for the project, was built with TinyVGG architecture as a reference. The project’s graphical interface was built with Pyglet, OpenGL bindings for Python. All of the code has been handwritten by me, Nick Kipshidze. I have not used search engines besides researching of the reference architecture or about the dataset. I have not used any code generating AI in my project, including ChatGPT. The project was uploaded on GitHub and you may have access to the repository here: https://github.com/nickkipshidze/handwritten-digits-reco
Project was built using Python and PyTorch as the primary framework. The dataset was handwritten by me consisting of 1640 handwritten Georgian letters. Took me 2 days to complete the whole dataset. The model was painstakingly trained over 10,000 epochs in Google's colab environment. The project’s graphical interface was built with Pyglet, OpenGL bindings for Python. All of the code has been handwritten by me, Nick Kipshidze. I have not used search engines besides researching of the reference architecture or about the dataset. I have not used any code generating AI in my project, including ChatGPT. The project was uploaded on GitHub and you may have access to the repository here: https://github.com/nickkipshidze/handwritten-letters-reco