Gautam Machiraju

Thesis Research

Intuitively, my goal is to work with AI to understand emergent phenomena or better inform our decision-making. To do this, we are pushing the needle on Explainable AI copilots capable of parsing unstructured scientific data for the discovery of class-salient regions of that data. Methodologically, I currently work in the intersection of (A) deep learning architecture development, (B) relational reasoning over long-range contexts, and (C) interpretability and concept discovery. Motivated by scientific and biomedical applications, my thesis work focuses on bettering our understanding of spatial systems and the key actors that distinguish such systems from one another. To this end, we seek to extract human-interpretable, differentially expressed regions captured in high-resolution, multiplexed imagery. My thesis aims include:

Designing explainability evaluation frameworks for architectural selection

For deployment in human-centered applications, evaluation of classifiers can't be anchored on accurate predictions alone. Are some architectures inherently better at detecting class-salient elements?

Designing inherently interpretable architectures & explanation methods

We reason that richer representations of data should construct more interpretable model explanations. For high-dimensional images, this involves leveraging the expressivity of Foundation Models and learning spatial context and long-range dependencies at multiple scales. Our architecture designs borrow ideas from systems theory, distributional semantics and language modeling, statistical inference, and graph theory.

Applications to spatial biology for biomarker discovery

We seek to improve cancer prognostics by discovering histo-molecular biomarkers of cancer progression in the tumor microenvironment. To do this, we analyze state-of-the-art highly multiplexed immunofluorescence tissue microscopy images (>35 image channels) through amazing collaborations at GE Global Research. In such application and data domains, classifier explainability is both necessary and opens up a new frontier in explanation mapping due to the increased channel dimensionality of multiplexed images.

Preprints & Publications

While my thesis research focuses on explaianble AI in the context of high-dimensional computer vision, I've been privileged to work in several methods and application domains, including: graph representation learning (for small molecule function prediction), ML for large-scale deployment (clinical decision support via the EHR, mobile health, global health monitoring via satellite imagery), factorization methods (multi-omics & wearables data integration), applied inference (to identify cancer-associated gene enhancers), and mathematical and physics-based modeling (of tumor growth and protein shedding). A copy of my CV can be found at the bottom of this page. Stay tuned for future work on:

1. Sept 2023: ion-binding site segmentation in protein structures driven by FM interpretability & explainability
2. Nov 2023: medical VLMs that pass medical board exams
3. Nov 2023: zero-shot semantic segmentation with long medical text support
4. Nov 2023: zero-shot relation extraction in video streams
5. Spring 2024: FMs trained on satellite imagery to predict measures of maternal health & child mortality
6. Spring 2024: FMs trained on satellite imagery to perform few-shot human labor trafficking encampments
7. Spring 2024: applying interpretability & explainability tools for in silico discovery of spatial biomarkers in multiplexed tissue imaging

• Selected
• All

Vitæ

More details (projects, collaborators, talks, academic service, relevant coursework) can be found on my CV and LinkedIn page.

• Stanford University 2018 - present

  Ph.D. Student
  Biomedical Data Science
  Instructive AI copilots for scientific discovery
• IBM Research Summer 2023

  Student Researcher
  Visually grounded Foundation Models for video
• Canary Center at Stanford Sep 2016 - Aug 2018

  Bioinformatics Research Assistant
  Mathematical modeling of tumors,
  NLP of biomedical literature,
  & anomaly detection in multi-omic time-series
• Strateos Summer 2016

  Bioengineering SDE Intern
  Software engineering for robotics:
  queueing, optimization, search
• Center for Computational Biology Sep 2015 - June 2016

  Undergraduate Researcher
  Genomic sequence alignment algorithms
• Strand Life Sciences Summer 2015

  Bioinformatics Intern
  Genomic rare variant identification
• Lawrence Berkeley National Lab July 2013 – May 2015

  Undergraduate Researcher
  Computational Biophysics &
  structural biology of enzymes
• Berkeley BioLabs Summer 2014

  Bioengineering Intern
  Gene transfer wet-lab experimentation
• University of California, Berkeley 2012 - 2016

  B.A. Student
  Applied Mathematics
  (emphasis in Mathematical Biology)
  Minor in Bioengineering

Process

One of my favorite aspects of research is thinking about aesthetic and design when communicating technical ideas. This drive to understand ideas by visually communicating them (often to myself) sparked as a dyslexic Maths undergraduate. Despite my numerous interests in Maths, I struggled to parse and conceptualize blocks of textual abstraction in modern mathematical presentation, typical of standard teaching materials. I thus relied heavily on intuition and visual proofs as mental anchors. Thanks in part to training as a CIR Scholar at Stanford's Hasso Plattner Institute of Design, I cartoon-ify almost everything I work on and often spend Friday afternoons reflecting on, mocking up, and refining any discussed concepts.

• Selected
• All

Detecting salient objects
toy examples to show class differential features, presented in behind our ICML IMLH 2023 paper

Systems for Foundation Models
describing advances on chip design + distributed training

New architecture to learn salient objects
depicting graphical data structures behind our ICML IMLH 2023 paper

Nash Equilibria for joint optimizers
multiple choice grid depicted for each training step

Mpox mobile surveillance
graphical abstract behind our Nature Medicine (2023) paper

Increasing context lengths
comparison of lab's architectures

Detecting salient objects
toy examples to show class differential features, presented in behind our ICML IMLH 2023 paper

Systems for Foundation Models
describing advances on chip design + distributed training

New architecture to learn salient objects
depicting graphical data structures behind our ICML IMLH 2023 paper

Nash Equilibria for joint optimizers
multiple choice grid depicted for each training step

Mpox mobile surveillance
graphical abstract behind our Nature Medicine (2023) paper

Increasing context lengths
comparison of lab's architectures

Evaluating model explanations
graphical abstract behind our ECCV 2022 paper

Vision for instructive AI
talking points for where models could help humans

Global health monitoring
graphical abstract behind remote sensing for health surveillance

Hallmarks of cancer
background information on cancer progression

Visual grounding
desiderata to create more expressive Foundation Models

Physics-based model of cancer
graphical abstract for our Nature Sci Reports (2021) paper

Acknowledgement