Mingi Kang

mkang2[at]bowdoin[dot]edu

mingikang31[at]gmail[dot]com

Hi! I am a recent grad from Bowdoin College with a major in Computer Science and a minor in Mathematics.

My research areas are in efficient deep learning and neural architecture design. I am particularly passionate about designing deep learning algorithms and architectures that prioritize both efficiency and interpretability. Some of my recent works include:

Convolutional Linear Attention (CLA)
Convolutional Nearest Neighbors (ConvNN)
Accelerated MRI Reconstruction using Supervised Learning
Integrated Gaussian Linear Unit (IGLU) Activation Function

Outside of research, I enjoy snowboarding, basketball, cooking, and exploring new cafes. I am also a proud brother to a medical student sister at USC Keck and a dedicated fan of the St. Louis Cardinals and LA Dodgers. Please feel free to reach out if you would like to connect!

selected publications

arXiv

Interpolation between Convolution and Attention via K-Nearest Neighbors

Mingi Kang

2026

arXiv Bib

@article{kang2026interpolation,
  title = {Interpolation between Convolution and Attention via K-Nearest Neighbors},
  author = {Kang, Mingi},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2606.14725},
}

ECML-PKDD
IGLU: The Integrated Gaussian Linear Unit Activation Function

Mingi Kang, Zai Yang, and Jeova Farias Sales Rocha Neto

2026

Abs arXiv Bib

Activation functions are fundamental to deep neural networks, governing gradient flow, optimization stability, and representational capacity. Within historic deep architectures, while ReLU has been the dominant choice for the activation function, modern transformer-based models increasingly are adopting smoother alternatives such as GELU and other self-gated alternatives. Despite their empirical success, the mathematical relationships among these functions and the principles underlying their effectiveness remains only partially understood. We introduce IGLU, a parametric activation function derived as a scale mixture of GELU gates under a half-normal mixing distribution. This derivation yields a closed-form expression whose gating component is exactly the Cauchy CDF, providing a principled one-parameter family that continuously interpolates between identity-like and ReLU-like behavior via a single sharpness parameter σ. Unlike GELU’s Gaussian gate, IGLU’s heavy-tailed Cauchy gate decays polynomially in the negative tail, guaranteeing non-zero gradients for all finite inputs and offering greater robustness to vanishing gradients. We further introduce IGLU-Approx, a computationally efficient rational approximation of IGLU expressed entirely in terms of ReLU operations that eliminates transcendental function evaluation. Through evaluations on CIFAR-10, CIFAR-100, and WikiText-103 across ResNet-20, ViT-Tiny, and GPT-2 Small, IGLU achieves competitive or superior performance on both vision and language datasets against ReLU and GELU baselines, with IGLU-Approx recovering this performance at substantially reduced computational cost. In particular, we show that employing a heavy-tailed gate leads to considerable performance gains in heavily imbalanced classification datasets.
@article{kang2026igluintegratedgaussianlinear, title = {IGLU: The Integrated Gaussian Linear Unit Activation Function}, author = {Kang, Mingi and Yang, Zai and Neto, Jeova Farias Sales Rocha}, year = {2026}, eprint = {2603.06861}, archiveprefix = {arXiv}, primaryclass = {cs.LG}, url = {https://arxiv.org/abs/2603.06861}, }

news

Jun 26, 2026	Awarded NSF NAIRR Pilot Compute Grant (NAIRR260070) for 2026-2027, serving as a co-PI/Project Admin with Professor Jeova Farias from Bowdoin College! This grant will support our research in efficient deep learning techniques, focusing on the development of Convolutional Linear Attention (CLA) and continuation of Convolutional Nearest Neighbors (ConvNN).
May 28, 2026	IGLU: The Integrated Gaussian Linear Unit has been accepted into the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) 2026!
May 23, 2026	Graduated from Bowdoin College with a Bachelor of Arts in Computer Science with honors and a minor in Mathematics! Honors Thesis: Interpolation between Convolution and Attention via K-Nearest Neighbors (Advisor: Professor Jeova Farias Sales Rocha Neto). Thesis PDF
May 01, 2026	Awarded the Computer Science Senior-Year Prize by the Department of Computer Science at Bowdoin College for achieving the highest distinction in the major program in computer science.
Mar 14, 2026	Awarded AI Grant from the Hastings Initiative for AI and Humanity at Bowdoin College to support advanced GPU computing training to extend Convolutional Nearest Neighbors (ConvNN) to CUDA/C++/Triton for faster inference and training.