I'm trying to optimize a notoriously complex loss function, navigating a high-dimensional, non-convex landscape—one small step at a time. Probably, so are you!
My learnings from this winding journey:
- Choose your objective with caution—you drift toward whatever you optimize.
- Curate what you train on—you learn from what you keep feeding yourself.
- Set the pace—bold enough to move, small enough not to overshoot.
- Favour signal over noise—small, informative steps beat thrashing.
- Rein in the extremes—when gradients explode, nobody benefits.
- Beware tunnel vision (overfitting)—a little regularization keeps you adaptable.
- Use momentum—keep going; let earlier progress carry you through flat stretches.
I hold an MEng in Electrical and Computer Engineering (2015, top 3% of my class) from Aristotle University of Thessaloniki, Greece, and an MSc in Artificial Intelligence (2022, magna cum laude) from KU Leuven, Belgium. I was fortunate to learn from outstanding professors and mentors in mathematics, physics, engineering, and computer science.
I have authored two papers:
- Antoniadis, I., Vercruyssen, V. and 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 (Nov 2015–Jul 2016). There, I contributed to an EU-funded H2020 project on cloud computing—when the field was still in its early stages—and worked with a large consortium of European institutions.
I continued as a software engineer at Veltio (Dec 2016–Jul 2018), an Oracle partner offering supply-chain automation solutions. I worked on real-world, large-scale problems alongside an exceptional team and led the development of data pipelines and systems used by major international retailers, including Sainsbury's in the UK.
I then joined the Intelligent Systems and Software Engineering Lab (ISSEL) at the Department of Electrical and Computer Engineering, AUTH, as an ML research engineer (Oct 2018–Sep 2021). I was the technical lead on an EU-funded project on energy monitoring and load disaggregation: applied ML research, NLP pipelines (e.g. BERT, topic models), and a high-throughput event streaming engine for real-time smart-meter analytics.
In September 2022 I joined Expedia Group in London as a machine learning scientist. On the Content & Relevance team I work on large-scale ranking and retrieval—reviews, amenities, and property understanding—using deep learning, LLMs, and multimodal methods. My recent work has included cross-brand review ranking, semantic relevance and distillation for low-latency embeddings, LLM-as-judge labelling, internal TensorFlow ranking frameworks shared across teams, distributed evaluation tooling, and research on bias, calibration, and data pruning.
*My first job was in 2011, during my second year at AUTH, as a part-time support representative at OTE, the largest telecommunications company in Greece.
I find it exciting to push human boundaries with technology, and I believe we have a responsibility to leave the world better for future generations.
All it takes is one small step at a time!
Past side projects, coursework, and research code live in separate repositories. Most of the older ones are archived on GitHub (read-only snapshots; not actively maintained) — the algorithm-visualiser portfolio below is the exception: it's active and still growing.
Algorithm visualisers (self-directed; from-scratch NumPy implementations, each with an interactive step-by-step Streamlit + Plotly walkthrough — no scikit-learn/PyTorch in the core algorithm)
Clustering
- kmeans-viz — k-means, manual centroid placement
- dbscan-viz — DBSCAN density clustering, core/border/noise points
- gmm-viz — Gaussian mixture models, EM algorithm
Dimensionality reduction
- pca-viz — PCA: covariance, eigenvectors, rank-k reconstruction
- umap-viz — UMAP: fuzzy simplicial graph + embedding SGD
- tsne-viz — t-SNE: KL-divergence descent on affinities
Classification & ensembles
- perceptron-viz — perceptron + SGD/momentum/Adam gradient descent
- svm-viz — soft-margin kernel SVM via SMO
- random-forest-viz — bagging, feature subsampling, OOB error
Deep learning building blocks
- backprop-viz — MLP backpropagation, gradient flow
- transformer-attention-viz — scaled dot-product self-attention, multi-head
Generative & self-supervised models
- vae-viz — variational autoencoder, reparameterisation trick
- diffusion-viz — DDPM forward/reverse diffusion process
- contrastive-learning-viz — SimCLR-style contrastive learning, NT-Xent loss
Graph algorithms
- dijkstra-viz — Dijkstra & A* pathfinding on a grid
- mst-viz — minimum spanning tree, Kruskal & Prim
Probabilistic methods, state estimation & signal processing
- mcmc-viz — Metropolis-Hastings MCMC sampling
- kalman-filter-viz — Kalman filter predict/update tracking
- fft-viz — Cooley-Tukey FFT, butterfly diagram
Reinforcement learning
- qlearning-viz — Q-learning vs SARSA on a stochastic grid world
Personal notebooks & experiments (self-directed; not part of a degree curriculum)
- bias-variance-decomposition — bias–variance decomposition
- fair-binary-classification — fairness on Adult (AIF360)
- gaussian-bandits — multi-armed bandits
- example-level-gradient-analysis — per-example gradients
- kepler-exoplanet-prediction — Kepler / KOI classification notebook
- sorting — Python sorting algorithms, pytest, benchmarks
KU Leuven — Master of Artificial Intelligence
Coursework
- cart — CART / decision trees
- grid-world-mdp — grid-world MDP, policy iteration
- taxi-rides-mapreduce — Hadoop MapReduce & Spark taxi analytics
- locality-sensitive-hashing — LSH on Stack Overflow posts
Thesis & published research
- cash-for-unsupervised-ad — Master’s thesis code extended to the LIDTA 2022 (ECML/PKDD) paper: CASH / AutoML for unsupervised anomaly detection
Aristotle University
Diploma thesis
- insight-qa — Semantic question answering (Java, Elasticsearch, LDA)
Coursework
- pagerank, octree-division — Parallel C (PageRank; octree spatial division)
You can find my full CV here.