Welcome on David Schulte

Customer return predictions

Sun, 08 Feb 2026 00:00:00 +0000

Motivation

Return rates in E-commerce are significantly higher than in retail. Especially when buying clothes online, customers expect free return shipping. This comes with immense economic as well as ecological costs.
Therefore, online retailers are interested in predicting customer returns at the time of their order. This enables them to better plan ahead. It also creates the possibility to subtly nudge users to not order a specific item they will return with a high probability.

Monitoring of the Cryptocurrency Transaction Graph

Sun, 08 Feb 2026 00:00:00 +0000

Motivation

In recent years, cryptocurrencies made their way into the mainstream. Their decentralized ungoverned nature made them attractive to people that were unsatisfied with the current financial system. Their extreme growth of value made them attractive to investors. Their volatility made them attractive to speculators. Their anonymity made them attractive to criminals.
How anonymous and decentralized the most prominent cryptocurrency bitcoin really is, was being researched resulting in surprising results. You can read more about it in the Section Further Research down below.

Reinforcement Learning for 3-player Chinese Checkers

Sun, 08 Feb 2026 00:00:00 +0000

Introduction

Motivation

Like for many aspiring Machine Learning enthusiasts, the development of AlphaZero could be described as the one event that made me sure, this is the right field for me. I was curious how a program can learn such sophisticated tactics in several domains just by playing against itself. I researched the design of its algorithm and was fascinated by it, I decided to dedicate my Bachelor’s thesis to experimenting with it.
The program is designed for learning a strategy for two-player zero-sum games. But what would happen, if we break this assumption?

Less is More: Parameter-Efficient Selection of Intermediate Tasks for Transfer Learning

Sat, 19 Oct 2024 00:00:00 +0000

Intermediate task transfer learning can greatly improve model performance. If, for example, one has little training data for emotion detection, first fine-tuning a language model on a sentiment classification dataset may improve performance strongly. But which task to choose for transfer learning? Prior methods producing useful task rankings are infeasible for large source pools, as they require forward passes through all source language models. We overcome this by introducing Embedding Space Maps (ESMs), light-weight neural networks that approximate the effect of fine-tuning a language model. We conduct the largest study on NLP task transferability and task selection with 12k source-target pairs. We find that applying ESMs on a prior method reduces execution time and disk space usage by factors of 10 and 278, respectively, while retaining high selection performance (avg. regret@5 score of 2.95).

Solar Forecasting

Tue, 01 Mar 2022 00:00:00 +0000

Motivation

Solar energy is one of the fundaments of our transition toward green energy. With all its advantages over other power sources, solar energy has one drawback that complicates its use: It only gets generated when it’s sunny. To utilize an efficient mix of energy sources, it is important to forecast their availability in the future.

Data

The data used is the daily solar power generation in Germany by its four network operators (50Hertz, Amprion, Tennet and Transnet BW). The data is provided by Fraunhofer Institute for Solar Energy Systems ISE and is publicly available on Energy-Charts.

About

Mon, 01 Jan 0001 00:00:00 +0000

Hi, my name is David Schulte. You can see my CV below.