Shusma Balaji

Master’s Project Dissertation - Awarded First Class

The ideation of this project originated from collaborating with the Neuroscientists at Krupic Lab from Cambridge University’s PDN Department, whose research focuses on the brain’s spatial navigation system, specifically the place and grid cells, neurons related to spatial memory and recognition. After evaluating the Lab’s research experiments, an idea was proposed to investigate Real-time Neural Decoding of rat’s behaviour from the… Show more

Master’s Project Dissertation - Awarded First Class

The ideation of this project originated from collaborating with the Neuroscientists at Krupic Lab from Cambridge University’s PDN Department, whose research focuses on the brain’s spatial navigation system, specifically the place and grid cells, neurons related to spatial memory and recognition. After evaluating the Lab’s research experiments, an idea was proposed to investigate Real-time Neural Decoding of rat’s behaviour from the Neural Response in the Hippocampus/Entorhinal Cortex (MEC) area.


1. Extracting Neural and Position Data (supplied by the Krupic Lab)
2. Initial Data Pre-processing (using Matlab)
3. Neural Decoders (using Python)

Traditional Decoders - Linear Regression, Non-linear Support Vector Regression (SVR) and Kalman Filters
Deep-Learning Decoders - Recurrent Neural Networks (Gated Recurrent Units GRUs and Long-Short Term Memory cells LSTMs) run on Google Colab’s Free GPU. Show less

Design Case Study (MEng Coursework Module) - Awarded First Class

I designed a device to monitor and manage stress in commercial and professional environments. This pitch deck details the product lifecycle from Ideation to Embodiment Design with specific focuses on what technologies and algorithms to leverage.

MEng Coursework Module (Awarded First Class)

Gaussian Processes | Probabilistic Machine Learning | Latent Dirichlet Allocation

The Allstate D3 Department also allotted Project Lightbulb time for personal development which I used to explore and understand Deep Learning. By presenting the findings from my investigation to members in the Data Science Team and seeking advice I was able to fuel further exploration. Firstly, I began by researching how to build Neural Networks for Deep Learning with TensorFlow. I experimented with building Word Embeddings with Word2Vec Continuous Bag Of Words & Skip-gram models for Sentiment… Show more

The Allstate D3 Department also allotted Project Lightbulb time for personal development which I used to explore and understand Deep Learning. By presenting the findings from my investigation to members in the Data Science Team and seeking advice I was able to fuel further exploration. Firstly, I began by researching how to build Neural Networks for Deep Learning with TensorFlow. I experimented with building Word Embeddings with Word2Vec Continuous Bag Of Words & Skip-gram models for Sentiment Analysis. I built Recurrent Neural Networks with LSTM (Long Short Term Memory) cells to perform Sentiment Classification on IMDB Reviews, then compared performance when using a pre-trained GloVe Neural Network. Played around with Optical Character Recognition using Bidirectional Recurrent Neural Networks on the MIT OCR Dataset. Show less