Aviraj Newatia
PhD student in Computer Science
University of Cambridge · CAMRAL
Hi, I’m Avi. I’m a PhD student in Computer Science at the University of Cambridge, in the Cambridge Resilient Autonomous Learning Group (CAMRAL), supervised by Dr Rika Antonova. My PhD is funded by ARIA’s Robot Dexterity programme as part of the Ada-GO project.
I work on simulators, physics-informed learning and robotics: how to adapt simulators, solvers and learned policies without breaking the dynamics they model. Before this I worked on machine unlearning with Dr Rahul Krishnan at the Vector Institute, and on quantum circuit design with Dr Hans-Arno Jacobsen’s group. I did my BSc in Computer Science and Mathematics at the University of Toronto.
Outside research you will usually find me running, swimming, reading, or writing. Occasionally I also like to go climbing, play tennis, squash, or some padel.
News
- Structural Affinity, an algorithm for coherent cage-based shape design, is on arXiv, with code.
- Co-design Gym, a benchmark for embodiment–policy co-optimisation, is on arXiv.
- Forget set-free unlearning (Reload) published at ICLR 2026.
- Received an NVIDIA Academic Grant for simulation and modelling research.
- Probabilistic design of parametrised quantum circuits is on arXiv.
Show older (6)
- Started my PhD at Cambridge with Rika Antonova, funded by ARIA's Robot Dexterity programme.
- Raft, Paper, Scissors — A foray into distributed systems through eventuaaal convergence.
- Graduated from the University of Toronto (BSc, Computer Science and Mathematics).
- Tabular unlearning without a forget set appeared at the TRL workshop, NeurIPS 2024.
- Self-learning Machines based on Hamiltonian Echo Backpropagation — A read of this paper on physical self-learning machines and their theorization
- Demoed Reload, selective data forgetting and replacement, at Vector Institute Demo Day.
Selected publications
Unlearning Tabular Data Without a "Forget Set"
PaperLatest writing
All writing →
Activity
Background
Experience 8
- 2025 – nowPhD researcherCage-based deformation for computational shape design, and Co-design Gym, in the ARIA-funded Ada-GO project.
- 2024 – 2026Research intern, then collaboratorForget set-free unlearning with gradient-based weight saliency and sparsity; led to the NeurIPS 2024 workshop and ICLR 2026 papers.
- 2024 – 2026Computer systems researcherEvolutionary architecture search for quantum ML circuits, and the effect of error mitigation on QML models.
- 2024Bioinformatics researcherText embeddings and a contrastive method to cluster Sequence Read Archive metadata.
- 2022 – 2024Machine learning researcherFine-tuned Llama-2 on play scripts; built a real-time VQGAN+CLIP canvas with Whisper voice control.
- 2022Data catalogue research assistantJMeter test suites that found pipeline bottlenecks; fixes cut request time from 18 s to 0.8 s.
- 2022NLP internUnsupervised multilingual intent clustering of customer-support queries.
- 2019 – 2021R&D internAssociation mining on 2M+ transactions and an RNN customer-feedback classifier.
Education 3
- 2025 – nowPhD, Computer ScienceSupervised by Rika Antonova. Funded by ARIA's Robot Dexterity programme.
- 2021 – 2025BSc (Hons), Computer Science and MathematicsDean's List all four years. Graduate courses in computational imaging, visual computing systems and deep learning theory.
- – 2021International Baccalaureate, 42/45
Talks 1
- Aug 2024Reload: selective data forgetting and selective data replacement
Service & leadership 4
- 2026 – nowComputer Officer, MCR
- 2025 – nowRower
- 2024 – 2025Director of Academic AffairsFounded UofT St George's first STEM undergraduate research journal and ran the undergraduate colloquium.
- 2024 – 2025Undergraduate member
