Research Papers

Papers I liked and why each one is worth reading.

Transformers & LLMs

Attention Is All You Need2017
Vaswani, Shazeer, Parmar et al.· NeurIPS
The Transformer paper — the attention mechanism that sits underneath every modern LLM.
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding2019
Devlin, Chang, Lee, Toutanova· NAACL-HLT
Bidirectional pre-training that changed how we think about transfer learning in NLP.

Deep Residual Learning for Image Recognition2016
He, Zhang, Ren, Sun· CVPR
Residual connections solved the vanishing-gradient problem and made very deep networks trainable.
You Only Look Once: Unified, Real-Time Object Detection2016
Redmon, Divvala, Girshick, Farhadi· CVPR
Framing detection as a single-stage regression problem — the paper that made real-time object detection practical.

Generative Adversarial Networks2014
Goodfellow, Pouget-Abadie, Mirza et al.· NeurIPS
The adversarial training framework behind most modern image generation. Worth re-reading for the theoretical framing.

CausalStock: Deep End-to-end Causal Discovery for News-driven Stock Movement Prediction2024
Li, Sun, Lin et al.· NeurIPS
A clean example of grounding ML in causal structure rather than just correlation — directly applicable to my adversarial safety research.

ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM2021
Campos, Elvira, Gómez et al.· IEEE Transactions on Robotics
The SLAM baseline I benchmark robot-perception work against.
Scalable Multi-Session Visual SLAM in Large-Scale Scenes with Subgraph Optimization2025
Pan, Li, Fan et al.· IEEE TPAMI
Multi-session visual SLAM using subgraph optimization — relevant to my robot-perception coursework and panoptic segmentation project.

Papers I liked and why each one is worth reading.

Attention Is All You Need2017
Vaswani, Shazeer, Parmar et al.· NeurIPS
The Transformer paper — the attention mechanism that sits underneath every modern LLM.
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding2019
Devlin, Chang, Lee, Toutanova· NAACL-HLT
Bidirectional pre-training that changed how we think about transfer learning in NLP.

Deep Residual Learning for Image Recognition2016
He, Zhang, Ren, Sun· CVPR
Residual connections solved the vanishing-gradient problem and made very deep networks trainable.
You Only Look Once: Unified, Real-Time Object Detection2016
Redmon, Divvala, Girshick, Farhadi· CVPR
Framing detection as a single-stage regression problem — the paper that made real-time object detection practical.

Generative Adversarial Networks2014
Goodfellow, Pouget-Abadie, Mirza et al.· NeurIPS
The adversarial training framework behind most modern image generation. Worth re-reading for the theoretical framing.

CausalStock: Deep End-to-end Causal Discovery for News-driven Stock Movement Prediction2024
Li, Sun, Lin et al.· NeurIPS
A clean example of grounding ML in causal structure rather than just correlation — directly applicable to my adversarial safety research.

ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM2021
Campos, Elvira, Gómez et al.· IEEE Transactions on Robotics
The SLAM baseline I benchmark robot-perception work against.
Scalable Multi-Session Visual SLAM in Large-Scale Scenes with Subgraph Optimization2025
Pan, Li, Fan et al.· IEEE TPAMI
Multi-session visual SLAM using subgraph optimization — relevant to my robot-perception coursework and panoptic segmentation project.