welcome. i'm a masters student in machine learning at Carnegie Mellon.
i like computing. i really like computing that approximates the complex real-world. here is my curriculum vitae.
let's get to know each other. please scroll down, or use the navbar above.
on mobile? got motion sickness? prefer retro? try the static version of the website .
In my free time, I'm a MS student at the MLD at SCS, CMU. I'm a part of Prof. Pradeep Ravikumar's group. These days, I mostly work on representation learning. My overarching agenda is to discover learning paradigms that are democratic & usable in some sense.
The guiding itch under my skin is caused by the observation that, whatever your opinion of emergent abilities in large models, training a human to do the same thing is much cheaper, and the inference is more reliable. (Before you say something about Navier-Stokes or ETH, try doing that with a single agent).
Earlier, I used to work on representation learning as it pertained to federated learning and robotics. I was an undergrad at IIT Kharagpur. In 2025, I got my B.Tech in ECE and M.Tech in Vision & Intelligent Systems (from the E&ECE department), and a minor in CS (from the CS&E department). I worked with Prof. Debashish Chakravarty's undergrad robotics group, AGV, and with Prof. Saumik Bhattacharya.
The NSF, Boeing and IIT KGP Foundation supported my summer research on vertical FL and greedy DL with Prof. David Inouye at Purdue, and on fast map matching and anomaly detection for Amelia at CMU's Robotics Institute.
I'm excited by the synergy between compositional generalization and decision-theoretic inference. I posit we've already seen the principle works: framing large models' posttraining as deciding upon a next token en route to a planning objective, and composing said plans from abstract chains of thought.
I think we should target compositional structure, e.g. by explicitly modeling experimental design/instrumental variables, interventionals between causal abstractions, etc. But, probability evaluation for the usual Empirical Bayes is expensive to calibrate (and outright intractable in continuous spaces) for neural networks' output distributions. I'm studying nonparametric and semiparametric inference tools, especially polynomial approximation and automated debiasing, around this. I am especially interested in Neyman orthogonality-like conditions for variational inference. I am also intrigued by normalizing flows, moreso flow maps, as expressive representations that allow exact likelihood computation. It remains to extend to tractable likelihood control.
On the decision-theoretic side, a recent interest is decision-centric exploration. What I described above could be considered an analog of exploitation, but spending computational time on exploration allows us to extract 'more' out of a given datapoint and model size. Unsupervised and heuristic acquisition, e.g. intrinsic motivation/empowerment at train time and looped computation at test time, has already been promising. I think a principled formalization of 'extracting more' naturally considers the downstream exploitation. I am currently interested in consequences for simulation-based inference (which naturally leads to popular definitions of world modelling), and LLM evaluation and finetuning.
Interested in my earlier work (mostly on MLSys and robot perception)? Consult my Google Scholar profile.
A list of recent projects here.
For samples of my code, check out my GitHub profile. Sadly most of it is in Python, but I can spin up a mean C++ or Haskell codebase too. I have recently adopted AI-assisted programming, and consider myself to be proficient at that too. I wager >75% of my commits since February 2026 have used some AI.
My e-mail is aviamala@iitkgp.andrew.co.cmu..comedu. I anticipate your hate mail.
My email and LinkedIn are probably the easiest way to get a hold of me.
I also have a Twitter (avi_amalanshu) but I don't really use it.
Check out my blog: malansh on Medium.
I'm always on the lookout for interesting puzzles and research problems, especially stuff that's interdisciplinary or niche (underappreciated). Let's talk if you have something interesting and can use my contributions.
I enjoy dispensing advice and mentorship (to the extent that I often do so unsolicited). So feel free to solicit if you're interested! Folks who have taken my advice are doing great and those who've ignored it are invariably suffering.