Raja Sunkara

I am a Ph.D. candidate in Computer Science at Missouri S&T, advised by Prof. Ardhendu Tripathy in the Statistical Machine Learning Lab. I work on sequential decision making: algorithms for black-box optimization and multi-armed bandits with provable guarantees, and applying them to real problems in machine learning, including large language models.

My recent work, Adaptive Partitioning Schemes for Optimistic Optimization (ICML 2025), learns the low-dimensional structure of a black-box function and uses it to search faster, with better regret bounds than SequOOL. The code is a pip-installable package (GitHub).

Before my Ph.D., I completed a thesis M.S. at Missouri S&T with Prof. Tie Luo on computer vision under adverse conditions. That work produced SPD-Conv (ECML PKDD 2022), a CNN building block for small objects and low-resolution images that is widely used in YOLO-based detectors, and YOGA (Pattern Recognition 2023), a lightweight object detector for edge devices.

I earned my B.Tech. in Aerospace Engineering from the Indian Institute of Technology Madras, with a thesis supervised by Prof. Ranjith Mohan. Between degrees I spent about four years in industry, as a data analyst, data scientist and research engineer at Enfrien Innovations, Agrometrics and Matdun Labs, and as an R&D data science intern at Alcon.

Research interests: black-box and Bayesian optimization · multi-armed bandits and pure exploration · reinforcement learning · efficient deep learning for vision and LLMs.

CV (PDF) · Google Scholar · GitHub

news

Nov 04, 2026	The YOGA code now runs on current PyTorch, with pretrained COCO weights in one line (`torch.hub.load("TrustAIoT/YOGA_object-detection", "yoga_s")`) and a Colab quick start.
Oct 16, 2026	Our ICML 2025 paper, Adaptive Partitioning Schemes for Optimistic Optimization, is now on arXiv, with code, a pip-installable package and a Colab demo.
May 01, 2025	Our paper Adaptive Partitioning Schemes for Optimistic Optimization was accepted at ICML 2025.
Dec 14, 2024	I will be attending the Neural Information Processing Systems (NeurIPS) conference from December 10 to December 15, 2024, to present my paper.
Jul 15, 2024	I attended the International Conference on Machine Learning (ICML) virtually in July 2024.

selected publications

ICML
Adaptive Partitioning Schemes for Optimistic Optimization

Raja Sunkara, and Ardhendu Tripathy

In Proceedings of the 42nd International Conference on Machine Learning, 2025

Abs arXiv Bib HTML PDF Code Website

Applications such as engineering design often require us to optimize a black-box function, i.e., a system whose inner processing is not analytically known and whose gradients are not available. Practitioners often have a fixed budget for the number of function evaluations and the performance of an optimization algorithm is measured by its simple regret. In this paper, we study the class of "Optimistic Optimization" algorithms for black-box optimization that use a partitioning scheme for the domain. We develop algorithms that learn a good partitioning scheme and use flexible surrogate models such as neural networks in the optimization procedure. For multi-index functions on an m-dimensional subspace within d dimensions, our algorithm attains \tildeO(n^-β/ d) regret, where β= 1 + \fracd-m2m-1, as opposed to \tildeO(n^-1/d) for SequOOL, a state-of-the-art optimistic optimization algorithm. Our approach is competitive across a wide range of numerical benchmarks. Additionally, we introduce weight quantization in a large language model as a novel task for black-box optimization.
@inproceedings{sunkara2025adaptive, title = {Adaptive Partitioning Schemes for Optimistic Optimization}, author = {Sunkara, Raja and Tripathy, Ardhendu}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, series = {Proceedings of Machine Learning Research}, volume = {267}, pages = {58029--58064}, year = {2025}, publisher = {PMLR}, }
Pattern Recognition
YOGA: Deep Object Detection in the Wild with Lightweight Feature Learning and Multiscale Attention

Raja Sunkara, and Tie Luo

Pattern Recognition, 2023

Abs DOI arXiv Bib Code

YOGA is a deep-learning object detector built to run on low-end edge devices while staying accurate. Its backbone, CSPGhostNet, learns half of each layer’s feature maps with ordinary convolutions and produces the other half with cheap linear operations, which roughly halves the number of filters. Its neck, AFF-PANet, fuses features across scales with attentional feature fusion instead of plain concatenation. On COCO-val and COCO-testdev, compared with more than ten state-of-the-art detectors, YOGA gives the best trade-off between model size and accuracy: up to 22% higher AP with 23–34% fewer parameters and FLOPs. It also runs in real time on an NVIDIA Jetson Nano.
@article{sunkara2023yoga, title = {{YOGA}: Deep Object Detection in the Wild with Lightweight Feature Learning and Multiscale Attention}, author = {Sunkara, Raja and Luo, Tie}, journal = {Pattern Recognition}, volume = {139}, pages = {109451}, year = {2023}, publisher = {Elsevier}, doi = {10.1016/j.patcog.2023.109451}, }
ECML-PKDD
No More Strided Convolutions or Pooling: A New CNN Building Block for Low-Resolution Images and Small Objects

Raja Sunkara, and Tie Luo

In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD), 2022

Abs DOI arXiv Bib Code

Convolutional neural networks lose fine-grained information through strided convolutions and pooling layers. This hurts badly on harder tasks such as low-resolution images and small objects. SPD-Conv replaces each strided convolution or pooling layer with a space-to-depth (SPD) layer, which rearranges every 2×2 block of pixels into channels without discarding anything, followed by a non-strided convolution. The block is general and drops into most CNN architectures. Applied to YOLOv5 (object detection) and ResNet (image classification), it clearly outperforms state-of-the-art models on low-resolution images and small objects.
@inproceedings{sunkara2022no, title = {No More Strided Convolutions or Pooling: A New {CNN} Building Block for Low-Resolution Images and Small Objects}, author = {Sunkara, Raja and Luo, Tie}, booktitle = {Joint European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD)}, series = {Lecture Notes in Computer Science}, volume = {13715}, pages = {443--459}, year = {2022}, publisher = {Springer}, doi = {10.1007/978-3-031-26409-2_27}, }