Adithya Giri
adithyag (at) berkeley (dot) edu
I
'
m
a
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
S
t
u
d
e
n
t
w
i
t
h
a
f
o
c
u
s
o
n
A
t
t
e
n
t
i
o
n
-
B
a
s
e
d
N
e
t
w
o
r
k
s
a
n
d
C
o
m
p
u
t
e
r
V
i
s
i
o
n
.


I
l
o
v
e
p
l
a
y
i
n
g
b
a
s
k
e
t
b
a
l
l
a
n
d
h
i
k
i
n
g
w
h
e
n
I
'
m
n
o
t
b
r
e
a
k
i
n
g
m
y
c
o
m
p
u
t
e
r
t
r
a
i
n
i
n
g
l
a
r
g
e
,
c
o
o
l
a
n
d
n
e
w
m
o
d
e
l
s
.


Publications
IBiT: Utilizing Inductive Biases to Create a More Data Efficient Attention Mechanism
Visualization of attention head activations from the IBiT paper
A low-rank matrix factorization mechanism that injects inductive biases into Vision Transformers, achieving +2% ImageNet accuracy over prior state-of-the-art models. Also introduces an interpretable framework for attention heads that reduces reliance on positional encodings and improves out-of-distribution generalization.
GithubPreprint

Projects
Seismo — 1st Place, AI/ML Track at DataHacks 2026
A multi-model pipeline that uses historical earthquake data to simulate earthquakes in Southern California, predict future collapse risk, and improve building structure through reinforcement learning. Built with PyTorch and FastAPI.
MoEPath: Multi-Token Expert Prediction
Ongoing research reducing mixture-of-experts VRAM requirements by up to 90% through pre-gating expert prediction, increased shared experts, and efficient offloading of routed experts to disk.
Using Adversarial Training to Create More Robust Planetary Classification using Flux
Example of an adversarial perturbation applied to stellar light curve data
Adversarial training with FGSM and PGD attacks on stellar light curve classifiers, improving model reliability from 50% to 90% with zero accuracy regression.
Carnatic Music Classification Using Transformers
Adithya Giri
Education
  • University of California, Berkeley (2025-2028)
    • B.S Electrical Engineering and Computer Science
    • Regents’ and Chancellor’s Scholar (top 1–2% of admitted students)
    • Coursework: Efficient Algorithms (CS 170), Data Structures (CS 61B), C and Assembly (CS 61C), Machine Learning (CS 189), Computer Vision (CS 180), Natural Language Processing (CS 183)
    • Machine Learning at Berkeley (MLAB)
Technical Skills
  • Languages
    • Python, Java, SQL, C/C++
  • ML Frameworks
    • PyTorch, FSDP, DeepSpeed, HuggingFace, TensorFlow, WandB
  • Tools & Systems
    • Linux/Bash, Docker, SLURM, Git, GCP, AWS
  • Libraries
    • NumPy, pandas, Matplotlib, OpenCV, SciPy
See my work experience
See my awards

© 2026 Adithya Giri. All Rights Reserved