Ivelina Mladenova

Hi you! I’m Ivelina, but most people call me Eve.

Research

My PhD sits where stochastic modelling meets machine learning. I work with correlated, high-dimensional volatility models such as Heston and Wishart processes, where the covariance structure itself evolves over time. Pricing under these models leads to high-dimensional PDEs and Fourier-based pricing operators, which classical tools (Monte Carlo, Fourier pricing, numerical PDE solvers) struggle to scale to. So I build neural approaches instead: Physics-Informed Neural Networks (PINNs) and Deep Galerkin Methods (DGM). Underneath it all is functional analysis, the framework for the operators and function spaces these problems live in.

Industry experience

Took an interruption from the PhD to complete the internship

• Blue Raven AI (Oct 2025 – Jun 2026: a 6-month internship, extended by 2 months)
  Researched systematic equity trading strategies on 1-minute data (2015–2024). There was no existing infrastructure, so I built it from scratch: data pipelines (including SQL to load the data), feature engineering and a historical simulation framework, then researched and backtested strategies on top of it. (Proprietary work, so no public code.)

Before the PhD

Projects and coursework from my BSc, MSc and Earth-i internship. All written before ChatGPT existed: just me, the documentation and a lot of StackOverflow, and I enjoyed every bit of it.

• MSc thesis (Imperial): network time series
  Tested the Generalised Network Autoregressive (GNAR) model on a network of macroeconomic variables to forecast inflation, in R. Code

• MSc coursework: particle filtering for stochastic volatility
  Notebook. Everyone is familiar with Kalman filters, but particle filters (Sequential Monte Carlo) are a much bigger field of research, and this was one of my favourite modules. This is where I met the curse of dimensionality: as the state dimension grows, particle weights collapse onto a single particle (particle degeneracy). I’m still curious about work combining probabilistic modelling with machine learning to fix this.

• BSc thesis: optimisation & parameter inference
  I really enjoyed learning about good optimisation techniques that optimise harder functions via simpler surrogate ones. I first worked through the examples in Kenneth Lange’s MM Optimization Algorithms and then Numerical Analysis for Statisticians (2nd edition), implementing some of them in Python: classical problems like optimising cos(x), a quadratic function and the sample median, then OLS and LAD regression on house price data, finding the optimal intercepts of a linear model that fits the data with the EM optimisation technique. I then used these techniques to find optimal parameters for a point process, which I studied from Daley & Vere-Jones, An Introduction to the Theory of Point Processes (Volumes I and II): simulating a Hawkes process and estimating its parameters with the Expectation–Maximisation (EM) algorithm. Code

• BSc coursework: time series analysis in R
  Code

Industry experience before the PhD

• Startup apprenticeship (accountant)
  At 18 I completed an apprenticeship as an accountant at a startup.

• Sainsbury’s data placement (BSc sandwich year)
  A 1-year data placement, using statistical methods such as hypothesis testing to measure how promotions shifted customer behaviour and which customer characteristics drove the response.

• Earth-i internship (2021): earth observation
  Gathered, cleaned and analysed Sentinel SAR radar images of copper smelters, trying my own clustering approaches (k-means) to detect changes in activity. Code