01
Interpretable
deep learning
Rebuilding iterative algorithms as trainable network layers, with an emphasis on parameter efficiency, convergence, and generalization.
PHD RESEARCHER · IIT BOMBAY
Jayaprakash
Katual.
Turning signals into insight.
Making learning interpretable.
I work at the intersection of deep learning, optimization, and signal processing — building models whose structure we can understand.
SIGNAL → STRUCTUREφ(x; θ)
xk+1 = Sθ(Wxk + By)
OPTIMIZATION, UNFOLDED.
↓ A little more about me⌖ Mumbai, India
INTERPRETABLE MACHINE LEARNING✳DEEP UNFOLDING✳SIGNAL & IMAGE PROCESSING✳INVERSE PROBLEMS✳
01 / ABOUT MEA QUESTION THAT DRIVES MY WORK
What if we could
learn more,
and understand why?
Electrical Engineering · IIT Bombay Mumbai, India
I’m a doctoral researcher in Electrical Engineering at IIT Bombay, advised by Prof. Satish Mulleti.
My research turns classical optimization algorithms into learnable architectures. Through deep unfolding, I study how we can retain algorithmic structure while learning from data — with applications in image restoration, sparse recovery, and sampling theory.
I’m also interested in the theory behind these models: their convergence, generalization, and the conditions under which efficient reconstruction is possible. My earlier work at NIT Hamirpur explored machine learning for thermal and visual image analysis.
Follow my academic journey∇Grounded in
optimization
optimization
φDesigned for
interpretability
interpretability
∞Driven by
curiosity
curiosity
↗Focused on
reconstruction
reconstruction
02 / RESEARCH DIRECTIONSWHERE THEORY MEETS LEARNING
Structure is
a powerful prior.
Learning from data, guided by the mathematics of the problem.
Interpretable
02
Signal & image
reconstruction
Solving inverse problems in image deblurring and sparse signal recovery, including learning from unpaired or limited training data.
03
Sampling &
adaptive models
Recovering bandlimited signals from modulo measurements and studying unfolded models that adapt to changing domains.
AN IDEA, MADE VISIBLE
From iteration
to architecture.
Each optimization step becomes a network layer. Move through the steps to see a noisy signal settle toward its underlying structure.
Illustrative visualization · not a trained research modelDEEP UNFOLDING
yx̂
Underlying signal Layer estimate
03 / PUBLICATIONSFull profile on Google Scholar
Ideas, put
into words.
Work spanning unfolded networks, image reconstruction, and thermal image analysis.
* Equal contribution. Preprints are labeled separately from journal and conference publications.
04 / THE JOURNEYALWAYS A STUDENT
From electrical engineering
to interpretable machine learning.
From electrical engineering to research at the intersection of optimization and machine learning.
View my full CV2023 — PRESENTCURRENT
Indian Institute of
Technology Bombay
Ph.D. · Electrical Engineering
Interpretable machine learning, deep unfolded networks, and inverse problems. Advised by Prof. Satish Mulleti.
Doctoral Researcher
National Institute of
Technology Hamirpur
M.Tech. · Signal Processing and Control
Dissertation: Thermal Image Analysis Using Deep Neural Networks. Research in thermal imaging, ensemble learning, and medical image analysis.
CGPA 8.50 / 10 · Dissertation grade A
Class Gold Medallist
Indira Gandhi Institute
of Technology, Sarang
B.Tech. · Electrical Engineering · BPUT
A foundation in electrical engineering and the mathematics of signals and systems.
CGPA 8.56 / 10
BEYOND RESEARCH
GATE 2020 qualified, Electrical Engineering
PG Training & Placement Representative, NIT Hamirpur
Departmental Master’s Programme Committee member, 2021–23
05 / MY TOOLKITTOOLS FOR THINKING & BUILDING
From the whiteboard
to the implementation.
Code & computation
Python · MATLAB · NumPy · SciPy
Learning frameworks
PyTorch · TensorFlow · Keras · scikit-learn
Research workflow
LaTeX · Git · Convex optimization
06 / LET’S CONNECT
Good research starts
with a conversation.
Interested in interpretable learning, signal reconstruction,
or a shared research question? I’d love to hear from you.