talkMay 2026
Invited speaker — MLSS Melbourne 2026
Lecturing at the Machine Learning Summer School, Melbourne (by invitation from Maincode).
Australia · Available for research collaborations
ArpitGarg
Senior ML / Research Engineer
LLMs · Multimodal AI · Computer Vision · Efficient Training
scroll
CVPR 2026 · TPAMI · NeurIPS · ECCV · WACV · TikTok · CSIRO
Australia · 2026◆
About
Research that ships.
From frontier models to film.
Senior ML engineer and published researcher specialising in large language models, multimodal AI, computer vision, machine unlearning, and efficient training systems. Co-founder of A2.AI (a2ai.com.au).
Recognised for shipping research-grade systems into production at scale — from TikTok's Trust & Safety MLLMs to VFX pipelines on Mad Max: Furiosa, Mortal Kombat II, Deadpool, Mickey 17, Sonic 3, Sinners, Michael and A Complete Unknown.
Inventor on a US provisional patent (attention mechanism) and a granted UK design patent (AI-Assisted Rural & Indigenous Healthcare Robot). Co-investigator on an A$2.1M grant powering frontier-scale training on 256× NVIDIA H200 GPUs.
- 140K+
- GitHub repo visits
- 256×
- H200 GPUs deployed
- A$2.1M
- Grant co-investigator
- 9
- VFX films shipped
- 10+
- Peer-reviewed papers
- 2
- Patents (US + UK)
“A datacenter hums at 256 GPUs. Somewhere inside, a model learns to forget.”
Experience
A research career, shipped to production.
Lab to live system, from defence imagery and indie VFX through TikTok-scale Trust & Safety MLLMs to frontier H200 training.
- May 2025 — Present
Research Fellow & Visiting Research Scientist
AIML, University of Adelaide · CSIRO · A$2.1M Grant Co-Investigator · Adelaide, Australia
- Co-investigator on an A$2.1M ResetData grant to train frontier-scale foundation models (language, multimodal, reasoning) on a 256× NVIDIA H200 cluster.
- Owned end-to-end training methodology, alignment, controllability research, and stability/throughput validation of the multi-million-dollar datacenter.
- Authored compute and memory-efficient LLM training that simultaneously reduces wall-clock time and peak GPU memory; covered by a US provisional patent.
- Lead research on machine unlearning, LoRA / PEFT, and stable vision-language alignment — accepted at CVPR 2026 (SineProject), multiple NeurIPS 2026 submissions, TPAMI under review.
- Joint appointment at CSIRO advising on responsible-AI and trustworthy LLM/MLLM systems.
LLMsMLLMsUnlearningLoRA / PEFTH200 - Oct 2024 — Present
Senior Machine Learning Engineer
TikTok · Trust & Safety Research · Australia
- Designed and shipped novel MLLM architectures for Trust & Safety: +2–3% AUC on business data, +5% additional lift via ensemble and distillation.
- Stack: SigLIP, CLIP, SAM, Co-DETR, DINOv2, ConvNeXt vision backbones paired with Phi, Gemma, Mistral LLMs.
- Owned retraining, evaluation, and deployment of production safety models with cross-functional engineering and product teams.
- Mentored junior engineers and drove research-engineering handoffs; co-authored top-tier peer-reviewed publications through sustained academic partnerships.
MLLMProductionSigLIPDINOv2Mentorship - Jun 2023 — Sep 2024
Machine Learning Developer (Research)
Rising Sun Pictures · VFX ML Research · Adelaide, Australia
- Designed a novel background-augmentation and occlusion-aware loss for deepfake pipelines, reducing FID by 15%.
- Integrated into shipping VFX workflows on Mad Max: Furiosa, Deadpool, Mickey 17, La Brea, Sonic 3, and Sinners.
- Shipped a production gaze-estimation model improving facial authenticity in VFX shots, reducing gaze error by 4%.
- Contributed to generative-AI pipelines across transformers, GANs, diffusion, super-resolution, Gaussian Splatting / NeRF, and video/audio synthesis.
VFXDiffusionGANsNeRFGaussian Splatting - Jan 2020 — Jun 2023
Machine Learning Researcher
Adelaide Business School & UoA · Applied CV & NLP for Market Intelligence · Adelaide, Australia
- Built CV/NLP market-intelligence systems covering 5,000+ companies; +12% sentiment accuracy and +7% data-driven decision quality.
- Designed an NLP expert-recommendation system with custom embeddings scaling to 200,000+ professional profiles.
- Architected an automated DL framework for security-patch classification across 10,000+ vulnerabilities.
NLPEmbeddingsMarket intelligence - May 2018 — Mar 2019
Applied ML Research Intern
DRDO & WESEE (Indian Navy) · Defence R&D · India
- Built CV / DIP algorithms for satellite-imagery analysis at >1M-image scale (DRDO).
- Delivered 20+ mission-critical algorithms improving real-time decision speed by 21% and operational efficiency by 25% for naval weapons systems (WESEE).
Satellite imageryDefenceDIP
Education
Nov 2021 — Jan 2025
Ph.D., LLMs / MLLMs, Generative AI & Computer Vision
Australian Institute for Machine Learning, University of Adelaide
Collaborations: University of Oxford, University of Surrey, Monash University
Jul 2019 — Jun 2021
M.S., Artificial Intelligence & Data Science
The University of Adelaide
Adelaide, Australia
Aug 2015 — Jun 2019
B.Eng., Computer Engineering
Rajasthan Technical University
India
“Every paper is an argument with the field. Every result is a vote.”
Publications
Selected research.
Top-tier peer-reviewed venues across LLMs, multimodal AI, computer vision, and machine unlearning.
CVPR 2026accepted
SineProject: Machine Unlearning for Stable Vision-Language Alignment
A. Garg, H. Saratchandran, S. Lucey
NeurIPS 2026in submission
STRIDE: Training-Free Diversity Guidance via PCA-Directed Feature Perturbation in Single-Step Diffusion Models
A. Yadav, A. Garg, T. D. Huy, L. Liu
arXiv
NeurIPS 2026in submission
LR-LoRA: Parameter-Efficient Fine-Tuning with Learnable Rank
A. Garg, S. Lucey, H. Saratchandran
NeurIPS 2026in submission
Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting
R. Xu*, A. Garg* (co-first), H. Saratchandran, S. Lucey
NeurIPS 2026in submission
Stable Forgetting: Bounded Parameter-Efficient Unlearning in LLMs
A. Garg, H. Saratchandran, R. Garg, S. Lucey
arXiv
TPAMIin submission
AEON: Adaptive Estimation of Instance-Dependent ID/OOD Label Noise for Robust Learning
A. Garg, C. Nguyen, R. Felix, Y. Liu, T.-T. Do, G. Carneiro
arXiv
IMAVIS 2025published
PASS: Peer-Agreement Based Sample Selection for Training with Instance-Dependent Noisy Labels
A. Garg, C. Nguyen, R. Felix, T.-T. Do, G. Carneiro
ECCV 2024published
Instance-Dependent Noisy-Label Learning with Graphical-Model-Based Noise-Rate Estimation
A. Garg, C. Nguyen, R. Felix, T.-T. Do, G. Carneiro
WACV 2023published
Instance-Dependent Noisy-Label Learning via Graphical Modelling
A. Garg, C. Nguyen, R. Felix, T.-T. Do, G. Carneiro
WACV 2021published
Per-VIS: Person Retrieval in Video Surveillance Using Semantic Description
P. Shah, A. Garg, V. Gajjar
Live · Google Scholar
Full publication record, citations & h-index.
Continuously updated with new accepted papers and preprints.
“On set, a face is rebuilt one pixel at a time. In a lab, so is reality.”
Filmography · VFX ML Research
Ninefeatures.
OneMLpipeline.
Generative-AI pipelines built at Rising Sun Pictures — deepfake, gaze, super-resolution, NeRF / Gaussian Splatting — integrated into shipping VFX workflows on:
2024 · Warner Bros.
Mad Max: Furiosa
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mortal Kombat II
VFX ML Research
Rising Sun Pictures
2024 · Marvel Studios
Deadpool & Wolverine
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mickey 17
VFX ML Research
Rising Sun Pictures
2024 · Paramount
Sonic the Hedgehog 3
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Sinners
VFX ML Research
Rising Sun Pictures
2026 · Lionsgate
Michael
VFX ML Research
Rising Sun Pictures
2024 · Searchlight Pictures
A Complete Unknown
VFX ML Research
Rising Sun Pictures
2023 · NBC
La Brea
VFX ML Research
Rising Sun Pictures
2024 · Warner Bros.
Mad Max: Furiosa
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mortal Kombat II
VFX ML Research
Rising Sun Pictures
2024 · Marvel Studios
Deadpool & Wolverine
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mickey 17
VFX ML Research
Rising Sun Pictures
2024 · Paramount
Sonic the Hedgehog 3
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Sinners
VFX ML Research
Rising Sun Pictures
2026 · Lionsgate
Michael
VFX ML Research
Rising Sun Pictures
2024 · Searchlight Pictures
A Complete Unknown
VFX ML Research
Rising Sun Pictures
2023 · NBC
La Brea
VFX ML Research
Rising Sun Pictures
2023 · NBC
La Brea
VFX ML Research
Rising Sun Pictures
2024 · Searchlight Pictures
A Complete Unknown
VFX ML Research
Rising Sun Pictures
2026 · Lionsgate
Michael
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Sinners
VFX ML Research
Rising Sun Pictures
2024 · Paramount
Sonic the Hedgehog 3
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mickey 17
VFX ML Research
Rising Sun Pictures
2024 · Marvel Studios
Deadpool & Wolverine
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mortal Kombat II
VFX ML Research
Rising Sun Pictures
2024 · Warner Bros.
Mad Max: Furiosa
VFX ML Research
Rising Sun Pictures
2023 · NBC
La Brea
VFX ML Research
Rising Sun Pictures
2024 · Searchlight Pictures
A Complete Unknown
VFX ML Research
Rising Sun Pictures
2026 · Lionsgate
Michael
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Sinners
VFX ML Research
Rising Sun Pictures
2024 · Paramount
Sonic the Hedgehog 3
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mickey 17
VFX ML Research
Rising Sun Pictures
2024 · Marvel Studios
Deadpool & Wolverine
VFX ML Research
Rising Sun Pictures
2025 · Warner Bros.
Mortal Kombat II
VFX ML Research
Rising Sun Pictures
2024 · Warner Bros.
Mad Max: Furiosa
VFX ML Research
Rising Sun Pictures
Title cards
Full credits on IMDb 
2024Warner Bros.
Mad Max: Furiosa
VFX ML Research
View on IMDb
2025Warner Bros.
Mortal Kombat II
VFX ML Research
View on IMDb
2024Marvel Studios
Deadpool & Wolverine
VFX ML Research
View on IMDb
2025Warner Bros.
Mickey 17
VFX ML Research
View on IMDb
2024Paramount
Sonic the Hedgehog 3
VFX ML Research
View on IMDb
2025Warner Bros.
Sinners
VFX ML Research
View on IMDb
2026Lionsgate
Michael
VFX ML Research
View on IMDb
2024Searchlight Pictures
A Complete Unknown
VFX ML Research
View on IMDb
2023NBC
La Brea
VFX ML Research
View on IMDb
“The frontier moves. Latest, here.”
Latest
What's happening now.
Recent talks, papers, awards, patents and grants.
patentApr 2026
UK Design Patent granted (No. 6520933)
AI-Assisted Rural & Indigenous Healthcare Robot — Class 24, Medical Equipment. UK Intellectual Property Office.
paperMar 2026
SineProject accepted at CVPR 2026
Machine unlearning for stable vision-language alignment. First-author work with Saratchandran and Lucey.
patentMay 2026
US Provisional Patent filed
Attention mechanism for neural networks — compute and memory-efficient LLM training, productionised in internal pipelines.
paperJan 2026
AEON submitted to TPAMI
Adaptive estimation of instance-dependent ID/OOD label noise for robust learning. arXiv:2501.13389.
paperDec 2025
PASS published in Image and Vision Computing
Peer-agreement based sample selection for training with instance-dependent noisy labels.
awardAug 2025
ICML 2025 Best Reviewer — Gold Award
Recognised among the top reviewers worldwide.
grantMay 2025
A$2.1M ResetData grant — investigator
Lead investigator for frontier-scale foundation model training on 256× NVIDIA H200 GPUs.
pressMay 2025
Joint appointment as Visiting Research Scientist at CSIRO
Advising on responsible-AI, trustworthy LLM/MLLM systems, and national-scale safety research.
pressOct 2024
Joined TikTok Trust & Safety Research
Senior ML engineer designing MLLM architectures for production safety models at scale.
paperSep 2024
ECCV 2024 — Instance-Dependent Noisy-Label Learning
Graphical-model-based noise-rate estimation; published at ECCV 2024 (Springer).
pressOngoing
Open-source impact — 140,000+ repo visits
Public ML projects across LLMs, attention mechanisms, noisy-label learning, and PEFT.
Writing
Long-form essays.
Visual deep dives into the papers and ideas shaping modern deep learning. Each post is a fully-rendered HTML reading experience.
Claude Code
June 2026~ 8 min
Wire Any Model Into Claude Code — Self-Hosted Routing
Three environment variables, no patching: point Claude Code at any Anthropic-compatible endpoint you host yourself. Route to a cluster (claude-glm) or an on-device model (claude-fable5) — plus how to name them in the /model picker and fix the connection failures you'll actually hit on self-hosted endpoints.
Explainer
June 2026~ 12 min
Mixture of Experts — A Hands-On Workbook
Mixture of Experts explained from scratch. Why the smartest AI models don't use their whole brain, what an 'expert' actually is, and how a tiny router decides who answers — an interactive workbook you can play with and break. No ML background needed.
Explainer
2026~ 15 min
The Window — Why Your Chatbot Slows Down
An interactive, jargon-free explainer of the context window: what the model can actually see, why doubling the conversation quadruples the cost, and what 'lost in the middle' really means.
Explainer
2026~ 15 min
The Harness — What Wraps the Brain
The model is only part of the story. An interactive tour of the six things that wrap a language model — turning a single chat turn into an agent that can read, search, run code, and act.
Attention
2026~ 90 min
The Attention Atlas
A field guide to every major attention mechanism, from the original Transformer to Flash Attention 4 — seven families, fifty mechanisms, one decade.
Long Context
May 2026~ 20 min
Lighthouse Attention — A Visual Deep Dive
How a hierarchical, parameter-free trick lets you pre-train Transformers on million-token contexts — and still get a fully dense model at the end.
Fine-Tuning
2025~ 25 min
LIMA & LESS — The Statistics of Selecting Training Data
LIMA argued a thousand carefully chosen examples can rival a million sloppy ones. LESS gave us the math. A complete recipe for data-efficient fine-tuning.
Subquadratic LLMs
2025~ 18 min
LoLCATs · A Technical Walkthrough
Attention transfer as feature-map distillation, LoRA as residual correction, and a block-wise schedule that makes 405B-parameter linearization tractable.
RL × LM
2025~ 8 min
POPGym, considered as a language model
What if partially-observable reinforcement learning is just next-token prediction wearing a different costume?
Patents
Inventions on file.
US Provisional
Attention Mechanism for Neural Networks
Compute & memory-efficient training of large language models.
Filed May 2026
UK Design (Granted)
AI-Assisted Rural & Indigenous Healthcare Robot
Class 24, Medical Equipment · UK Intellectual Property Office
No. 6520933 · 29 April 2026
Honors & Recognition
On the record.
Award
ICML 2025 Best Reviewer — Gold Award
Recognised among the top reviewers worldwide.
Award
Investigator, A$2.1M ResetData Grant
Training large foundation models on 256× NVIDIA H200 GPUs.
Award
Invited Speaker, MLSS Melbourne 2026
By invitation from Maincode.
Award
Open-Source Impact
140,000+ visits across public ML repositories.