Research in Progress
My ongoing work covers three main areas:
1. Abstract Reasoning Corpus
I am conducting a review of existing methods for solving abstract reasoning tasks. These tasks test a model's ability to handle logical deduction, pattern recognition, and rule application. They go far beyond simple memorization. In this review, I group the main approaches into families, such as symbolic program synthesis, search-based algorithms, and neural networks that infer hidden rules from small amounts of data. My aim is to find out which strategies work best on unfamiliar problems, to identify where they commonly fail, and to highlight gaps that remain unexplored. The final output will be a useful guide for researchers who want to benchmark and improve reasoning abilities in future AI systems.
2. Financial Markets and AI
I use deep learning models to analyze financial time-series data and forecast market behavior. I rely on transformer and recurrent neural networks to capture complex patterns in asset returns, especially during periods of high economic uncertainty. In addition to standard price data, I also include alternative text sources, such as central bank statements and earnings call transcripts, to strengthen my predictions. I evaluate these models not only for their accuracy but also for their interpretability. I want to ensure that the patterns they learn are economically meaningful and can be trusted by practitioners for risk management and investment decisions.
3. The Future of Work
I study how generative AI is changing jobs and wages across different occupations. Using detailed task-level data from sources like O*NET and the BLS, I measure how much different skill levels and job functions are exposed to automation. I aim to separate tasks that are likely to be automated from those that will benefit from AI as a helpful tool. This allows me to project possible shifts in wage inequality and to identify job traits, such as creativity, complex problem-solving, and interpersonal interaction, which make them more resistant to automation. My findings are meant to support workforce training policies and broader discussions about the future distribution of income.