CCS335 _ Neural Networks and Deep Learning Laboratory_Lab Complete Record
Resume
1. Akshay Bhatia
Junior year engineering undergraduate
Bioinformatics and Systems Biology enthusiast
akshaybt@iitk.ac.in | +917318018788 | LinkedIn: akshay | Github: akshay
H-216/5, IIT Kanpur, Kanpur-208016
EDUCATION
Indian Institute of Technology, Kanpur Kanpur, India
Bachelor of Technology in Biological Sciences & Bioengineering Expected July 2020
Army Public School, Dhaula Kuan New Delhi, India
High School, CBSE; 93.8% (Percentile 98.3th) May 2015
Army Public School, Dhaula Kuan New Delhi, India
Middle School, CBSE; CGPA: 10.0/10.0 May 2013
TECHNICAL SKILLS
• Languages: Python, C, C++, Matlab, R
• Technologies: AWS, Git, Docker, Jekyll, LATEX
• Libraries: TensorFlow, PyTorch, Keras, Scikit-Learn, Numpy, Pandas, Jupyter, OpenCV, CUDA, ROS
RESEARCH EXPERIENCE
Prediction of PPIs from mRNA expression data April, 2019
Prof. R Sankararamakrisnan IIT Kanpur
◦ Used ARCHS4 AWS pipeline to fetch gene counts of human and mouse RNA-seq experiments from GEO and SRA databases.
◦ Also fetched raw gene sequences from the same databases to construct gene expression matrix.
∗ From the constructed expression matrix, 10,000 samples were randomly selected to construct gene expression correlation matrices
for mouse and human separately.
∗ From the extracted expression matrices, all pairwise gene correlations were calculated. For each gene set gjε GS and each gene gi
the mean correlation of the genes in the gene set to gi was calculated.
∗ Hence, the resulting gene set membership prediction matrix GMεM × N for M genes and N gene sets is generated using the mean
correlation.
∗ Row vector was then sorted and used to compute the AUC from the cumulative sum of −→si using trapezoidal integration.
◦ Comparison with existing PPI databases
∗ To predict PPIs, the three PPI networks, hu.MAP, BioGRID, and BioPLEX, are first converted to a gene set library as described by
Maayan et al (Nature 2018).
∗ Comparison of the predicted interactions was done using the IntAct database of PPIs.
Monte Carlo study of random polynomials to examine root behaviour Nov, 2018
Statistical Simulation (Prof. Subhajit Dutta) IIT Kanpur
◦ Used Monte Carlo simulations to find out behaviour of roots of random polynomial in complex plane.
◦ Random polynomials were generated from uniform discrete, standard Gaussian, standard Cauchy and standard exponential
distributions.
∗ Complex roots of the polynomial with iid coefficients tend to concentrate around the unit circle.
∗ If the coefficients are simulated from a heavy tailed distribution (standard cauchy), the roots cluster around on the circumference
of two concentric circles.
∗ The result was proven to be consistent with the theoretical results where the radii are given by (
ξ0
ξτ
)
1
τ and (
ξτ
ξn
)
1
n−τ , where ξτ
represents the coefficient with largest modulus, and n is the degree of random polynomial.
ACADEMIC & CO-CURRICULAR ACHIEVEMENTS
◦ 97.7th percentile among 150,000 students in JEE Advanced 2016
◦ 99.5th percentile among 1.2 million students in JEE Main 2016
◦ Awarded 5 star badge in 2015 for academic excellence for 5 consecutive years
◦ Selected as a Ratti Chattr Scholar (only 30 students) across all IITs by Panasonic, India in 2016.
2. SELECTED PROJECTS
Identifying Granger Causality in electroencephalography (EEG) data Nov, 2018
Time Series Analysis (Prof. Amit Mitra) IIT Kanpur
∗ The given dataset contained data for 10 alcoholic and 10 control subjects, with 10 runs per subject per paradigm.
∗ Data from 63 electrodes corresponded to 630 time series each containing 256 readings in microvolts.
· Constructed 10 Horizontal Visibility Graphs (HVGs) for each of the 63 channels. For each channel, a significant number of
features are extracted from its HVG.
· For each channel, extracted features are evaluated with Kruskal Wallis Test for alcoholic and control class attributed to the
selected channel. Top 3 abnormal channels were identified.
· For validation, SVM algorithm achieved an accuracy of 87.8% on the test data set based on three channels alone, which is a
strong indication of abnormality.
· Granger Causality was calculated for these three channels after model selection by Bayes Information Criterion (BIC).
Audio Data Mining Oct, 2018
Data Mining, Group project (Prof. Arnab Bhattacharya) IIT Kanpur
∗ Provided a different approach to song popularity prediction by not using the top popularity charts like US Top 40, Billboard, iTunes
sales, YouTube plays.
· Total of 518 audio features extracted using librosa python library, we removed the highly correlated. Used multi-relief algorithm
to find the top 100 features. Also extracted audio features provided by Echonest (now Spotify) for a set of 13,129 tracks.
· To get temporal features from complete audio file, we sampled 10 data points(frequency) from each second stored them in a
vector. By using the sampled data, we created a visibility graph for each audio file. In addition, also included spectrogram and
chromogram plots for image processing pathway.
· Obtained an accuracy of ∼40% using CNNs, 32% using Visibility Graphs, and ∼30% using generic audio features in the
prediction of song popularity.
Correct systematic heading drift using dual foot mounted configuration June, 2018
IoT System Design (Prof. Amey Karkare) IIT Kanpur
∗ Learned about Zero Velocity Update (ZUPT) algorithm for inertial sensing.
∗ The drawbacks of the existing single foot-mounted ZUPT-aided INS is the Systematic Heading Drift.
· Programmed ESP8266 with the help of Arduino UNO to configure it for connecting with iitk network. The code was provided by
instructor and bootloaded into the ESP8266 flash memory.
· Applied Kalman filter for data fusion from Inertial Measurement Units in the AWS cloud.
RELEVANT COURSEWORK
Introduction to Computing Probability & Statistics Molecular and Cell Biology Statistical Simulation Data Mining
Time Series Analysis IoT system design Bioinformatics & Computational Biology Probability Theory Linear Algebra
Information Theory Structural Biology Topics in Linear Programming Dynamic Modelling in Systems Biology
Italics signifies courses in upcoming semester
POSITION OF RESPONSIBILITIES
∗ Senior Executive, Show Management at Antaragni 2018, annual Cultural festival of IIT Kanpur
∗ Maintaining blog about social topics at humane-hamilton.netlify.com