Aadit Deshpande's repositories
u-nets-implementation
Implemented the U-Net architecture proposed by Ronneberger et. al, and used it for water body detection in a dataset of 2841 images from the Sentinel-2 satellite.
procedurally-generated-terrains
Adventures in procedurally generated terrain, using the Unity engine. Used Perlin Noise with octaves to create a semi-realistic landscape.
Aadit3003.github.io
Personal Website
colour-palette-extraction
Extracted 5-colour palettes from a dataset of 541 images of Fall 2021 Couture Fashion shows and visualized them using unsupervised learning algorithms.
Diabetic-Retinopathy_Autoencoder
Autoencoder-based Feature Selection for the SN_DREAMS diabetic retinopathy dataset. (With Prof. S. Raman)
mCNV-image-analysis
(CS F266) An ImageJ macro for Quantitative Analysis of OCTA images.
split-learning-healthcare
Split learning for privacy-preserving healthcare, and threats and defensive techniques for decentralized learning. (with Prof. Vinay Chamola)
CNNs-course-exercises
My Lab Assignments for the "CNNs" course by Andrew Ng (Coursera)
Compiler-Construction
(CS F363) Assignment as part of the Compiler Construction course, under Prof. Vandana Agarwal
epitran-aadit
A tool for transcribing orthographic text as IPA (International Phonetic Alphabet)
fst-noun-verb-sw-so
MorphoPhonological analysis of Somali and Swahili Noun + Verb forms
g2p-estonian-epitran
Estonian Support for Epitran
GoogLeNet-implementation
Implementing the GoogLeNet architecture proposed by C Szegedy et. al
llm-medical-personas
Examination of whether LLMs can maintain consistency over extended multiple text generation for 10 medical personas. 5 novel plausibility metrics proposed, and an ontology of common LLM errors.
llm-rhyme-eval
English and Dutch rhyming datasets (5k word pairs each) for five types of rhymes. Three open-source LLMs (Llama2, Llama3, CrystalChat) are tested on these datasets, with prompt variation.
neural-reinflection-paradigm-completion
Paradigm Completion and Reinflection for Kabardian (kbd), Swahili (swc), and Mixtec (xty).
s2st-cascading-e2e
A comparison of E2E and Cascading S2ST systems on the CVSS-C Spanish to English dataset (CommonVoice 4.0)
subword-tokenizers-low-resource
Subword Tokenizers for Shipibo-Konibo (shp) and Rarámuri (tar) to beat the Morfessor and Sentencepiece baselines
Twitter-Sentiment-Polarity-Analysis
(CS F469) Twitter Sentiment Polarity Analysis using Naive Bayes
genai-multilingual-tti
Training our own Multilingual Stable Diffusion model (French and German) on subsets of the WIT and CC-12 datasets.
github-readme-stats
:zap: Dynamically generated stats for your github readmes