Mert Kilickaya

Hello! I am an AI researcher specializing in deep learning for self-supervised visual representations. My work focuses on developing autonomous systems that continuously improve by extracting supervision from dynamic data.

I have contributed to self-supervised continual learning at the Learning to Learn Lab and earned my PhD in deep vision at the QUvA Lab under Arnold Smeulders.

Email | CV | Google Scholar | LinkedIn

News

Experience

Agendia
Agendia, Amsterdam, Netherlands
2023-Now

ML Researcher

TU Eindhoven
Learning to Learn Lab, Eindhoven University of Technology, Netherlands
2022-2023

ML Researcher

QUvA Deep Vision Lab
QUvA Deep Vision Lab, University of Amsterdam, Netherlands
2017-2022

PhD Researcher

Huawei Visual Search Lab
Huawei Visual Search Lab, Helsinki, Finland
2021 March - 2022 January

Research Scientist Intern

Research

HyTAS
HyTAS: A Hyperspectral Image Transformer Architecture Search Benchmark and Analysis
Fangqin Zhou, Mert Kilickaya, Joaquin Vanschoren, Ran Piao
ECCV 2024
arXiv GitHub

We present HyTAS, a quick method for hyperspectral vision transformer architecture search.

HyLITE
Locality-Aware Hyperspectral Classification
Fangqin Zhou, Mert Kilickaya, Joaquin Vanschoren
BMVC 2023
arXiv GitHub

We propose HyLITE, a vision-transformer for hyperspectral image classification.

Towards Label-Efficient Incremental Learning: A Survey
Towards Label-Efficient Incremental Learning: A Survey
Mert Kilickaya, Joost van de Weijer, Yuki Asano
Preprint 2023
arXiv Slides GitHub

We survey semi-, few-shot, and self-supervised incremental learning.

Are Labels Needed for Incremental Instance Learning?
Are Labels Needed for Incremental Instance Learning?
Mert Kilickaya, Joaquin Vanschoren
CVPRW 2023 (Oral)
arXiv Slides

We introduce VINIL, a self-supervised incremental instance learner.

Human-Object Interaction Detection via Weak Supervision
Human-Object Interaction Detection via Weak Supervision
Mert Kilickaya, Arnold Smeulders
BMVC 2021
arXiv Slides

We introduce Align-Former to detect human-object interactions from an image.

Structured Visual Search via Composition-aware Learning
Structured Visual Search via Composition-aware Learning
Mert Kilickaya, Arnold Smeulders
WACV 2021
arXiv Slides

We introduce composition-aware learning to improve visual image search.

Re-evaluating Automatic Metrics for Image Captioning
Re-evaluating Automatic Metrics for Image Captioning
Mert Kilickaya, Aykut Erdem, Nazli Ikizler-Cinbis, Erkut Erdem
EACL 2017
arXiv Slides

We propose Word-Mover Distance to evaluate image captioning.

Patents

Visual Image Search via Conversational Interaction (Huawei)

Mert Kilickaya, Baiqiang XIA

WO Patent, 2022

Network for Interacted Object Localization (Qualcomm)

Mert Kilickaya, Arnold Smeulders

US Patent, 2022

Subject-Object Interaction Recognition Model (Qualcomm)

Mert Kilickaya, Stratis Gavves, Arnold Smeulders

US Patent, 2022

Context-driven Learning of Human-object Interactions (Qualcomm)

Mert Kilickaya, Noureldien Hussein, Stratis Gavves, Arnold Smeulders

US Patent, 2021

Supervision

I am very fortunate to have crossed paths with these talented individuals.

Evangelos Tsimpouris
Evangelos Tsimpouris (MSc, TU/e)
2023-2024
Learning to Transfer for Continual Learning
Fangqin Zhou
Fangqin Zhou (PhD, TU/e)
2022-2024
Locality-Aware Vision Transformers
Ran Piao
Ran Piao (MSc, TU/e)
2022-2023
Rapid Transformer Architecture Search without Training
Tommie Kerssies
Tommie Kerssies (MSc, TU/e)
2022-2023
Test Time Adaptation
Ceren Gok-Yildirim
Ceren Gok-Yildirim (PhD, TU/e)
2022-2023
Learning to Adapt for Continual Learning
Kishan Parshotam
Kishan Parshotam (MSc, UvA)
2020-2021
Continual Learning of Object Instances
Tarun Krishna
Tarun Krishna (MSc, UvA)
2019-2020
Learning to rotate, scale and shift visual objects via disentanglement