Trying to optimize a notoriously complex loss function by navigating an extremely high-dimensional, non-convex space-time manifold, one small step at a time.
Most probably, so are you!
My learnings from this mysterious journey:
- Wisely choose the loss function(s) you want to optimize.
- Carefully choose your training data. This is what you'll be learning from.
- Adapt your learning rate accordingly; be careful not to overshoot.
- Make informative, small steps towards the right direction.
- Trust the power of the gradient but also enable clipping to keep it from exploding.
- Stay alert for you might overfit. Perhaps use some regularization.
- Don't expect to find any global minima.
- Escape the plateaus; you'll encounter plenty; like, a lot!
- Keep moving downhill!
I hold a diploma in Electrical and Computer Engineering (2015) and an advanced Master's degree in Artificial Intelligence (2022) from KU Leuven, Belgium. I was fortunate to have been educated by amazing professors, experts and mentors in the topics of mathematics, physics, engineering and computer science.
I have authored two papers:
- Antoniadis, I., Vercruyssen, V. & Davis, J.. (2022). Systematic Evaluation of CASH Search Strategies for Unsupervised Anomaly Detection. Proceedings of the Fourth International Workshop on Learning with Imbalanced Domains: Theory and Applications, in Proceedings of Machine Learning Research 183:8-22.
- I. I. Antoniadis, K. C. Chatzidimitriou and A. L. Symeonidis, "Security and Privacy for Smart Meters: A Data-Driven Mapping Study," 2019 IEEE PES Innovative Smart Grid Technologies Europe (ISGT-Europe), Bucharest, Romania, 2019, pp. 1-5, doi: 10.1109/ISGTEurope.2019.8905611.
My professional career started in 2015*, when I joined the Centre for Research and Technology Hellas (CERTH) as a research associate. There, I worked on a EU-funded project on cloud computing, when the field was still at its very early stages, and collaborated with a large consortium of european institutions.
I continued my journey as a software engineer at Veltio, a company (Oracle golden partner) offering supply-chain automation solutions. I worked on real-world, large-scale problems alongside an exceptional group of colleagues. Together we built data processing algorithms and automation systems that were used by major international actors, such as Sainsbury's, that drove significant growth and revenue.
I later joined the Intelligent Systems and Software Engineering Labgroup (ISSEL) of the Electrical and Computer Engineering department, AUTH, as a research associate, where I was responsible for the development and technical management of an EU-funded project related to energy monitoring and load disaggregation in consumer installations.
In July 2022 I joined Expedia Group, London, as a machine learning scientist. Being a member of the Relevance team, I work on xlarge-scale ranking problems using deep learning. I also devote a percentage of my time to impactful state-of-the-art deep learning research, trying to identify potential opportunities and incorporate/customize novel solutions to facilitate the traveler's journey.
*Actually, my first job was in 2011 during my second-year studies at AUTH, where I worked as a part-time support representative at OTE, the largest telecommunications company in Greece.
I find the idea of pushing the human boundaries using technology exciting and I believe in the responsibility of making the world a better place for future generations.
All it takes is one small positive step at a time!
You can find my full cv here.