CakeResume Job Search
Use the APP to find jobs.
Ranking: 7/646(Top 2% in the world)
It is a data science competition. In this image classification task, we get 42 categories of product images. Apply One-Cycle policy theory to find good learning rate, also implement augmentation to increase amount of training set. Use data pipe function in TF2.0 and Keras on model training and TPU to accelerate training processing.
Develop time series anomaly detection. Take event log data as training set, then preprocess raw data by transforming them into sequences and standardization. Build LSTM autoencoder to deal with sequence data. Compare Mean Absolute Error between input and reconstructed data to define the threshold to figure out the unusual sequences. Finally, collaborate with product team to apply the model on a dashboard.
Thesis: Photoinduced gold nanostructures using DNA biomaterial and the application in SERS