Pinned Repositories
alienInvasion
A ship shooting game, developed using pygame.
BNU-thesis-template
This is a simplified LaTex template of BNU thesis without cover. It uses .bib / biblatex to manage references in GBT 7714 style.
cart
An implementation of cart (Classfication and Regression Tree), with a queue rather than a recursive method
CNN-ViT
Implements MobileViT and ViT from scratch, compatible with graph execution mode in tensorflow. Compares multiple cnn and transformer-based models including VGG, ResNet, ViT and MobileViT on a small data set.
info-retrieval
This project implemented three tradional information retrieval systems - TF-IDF Vector Space Based, BM25, and Likelihood Based, as well as three learning-to-rank systems - Logistic Regression Model, LambdaMART, and MLP.
market-regime
Regime identification in financial markets using CNN, Correlation, K-means++, based on multivariate time series.
Prune-Finetune-LLM
Different pipelines which integrates Wanda pruning and LoRA fine-tuning of llama-7B, focusing on effects of the order of these two different approachs.
SelfPretrain-SegFinetune
Pre-text pre-training -> image segmentaion pipelines. Utilize contrastive learning and ViT as Encoder. Studied the effects of dataset sizes, dataset similarity and effects of fine-tuning.
VGGSPP
This repo implements a VGG model with SPP layer to enable the model to receive images with different shapes - gray images with the same shape, gray images with different shapes, bgr images with the same shape, bgr images with different shapes.
ViT-Mixup
This repo implements a ViT based model with Mixup Data Augmentation method. All the models including ViT are implemented from scratch using tensorflow
kangchengX's Repositories
kangchengX/alienInvasion
A ship shooting game, developed using pygame.
kangchengX/BNU-thesis-template
This is a simplified LaTex template of BNU thesis without cover. It uses .bib / biblatex to manage references in GBT 7714 style.
kangchengX/cart
An implementation of cart (Classfication and Regression Tree), with a queue rather than a recursive method
kangchengX/CNN-ViT
Implements MobileViT and ViT from scratch, compatible with graph execution mode in tensorflow. Compares multiple cnn and transformer-based models including VGG, ResNet, ViT and MobileViT on a small data set.
kangchengX/info-retrieval
This project implemented three tradional information retrieval systems - TF-IDF Vector Space Based, BM25, and Likelihood Based, as well as three learning-to-rank systems - Logistic Regression Model, LambdaMART, and MLP.
kangchengX/market-regime
Regime identification in financial markets using CNN, Correlation, K-means++, based on multivariate time series.
kangchengX/Prune-Finetune-LLM
Different pipelines which integrates Wanda pruning and LoRA fine-tuning of llama-7B, focusing on effects of the order of these two different approachs.
kangchengX/SelfPretrain-SegFinetune
Pre-text pre-training -> image segmentaion pipelines. Utilize contrastive learning and ViT as Encoder. Studied the effects of dataset sizes, dataset similarity and effects of fine-tuning.
kangchengX/VGGSPP
This repo implements a VGG model with SPP layer to enable the model to receive images with different shapes - gray images with the same shape, gray images with different shapes, bgr images with the same shape, bgr images with different shapes.
kangchengX/ViT-Mixup
This repo implements a ViT based model with Mixup Data Augmentation method. All the models including ViT are implemented from scratch using tensorflow
kangchengX/wanda
A simple and effective LLM pruning approach.