Harnessing AI for Image Recognition in Finance - the JP Morgan journey

In the continuously evolving landscape of technology and finance, JPMorgan Chase & Co. has embarked on a pioneering venture by integrating artificial intelligence (AI) in a novel and groundbreaking manner. Spearheaded by AI expert Manuela Veloso, the company has embraced image recognition technology, moving beyond the conventional detection of cats, dogs, or oranges, to something much more complex and potentially profitable: buy/sell trading decisions.

The Beginning of a Revolution in Trading

The journey into this new territory for JPMorgan commenced with the appointment of Veloso as the head of its AI research group in 2018. Veloso, who balances her role at JPMorgan with her position as the head of machine learning research at Carnegie Mellon University, introduced a fresh perspective. Her expertise in training AI systems in the more traditional fields of image recognition laid the groundwork for this innovative challenge.

A Fresh Perspective on the Trading Floor

Veloso's initial visit to the JPMorgan trading floor was a defining moment. Faced with an array of screens and a flurry of activity, she pondered over how humans digest the vast amount of information displayed and make swift, effective trading decisions.

This contemplation led to an epiphany. Veloso recognised that just as humans derive trading decisions from visual cues on their screens, machines could similarly be trained to interpret these images.

Conventionally, finance has relied on analysing numerical time series data. Veloso proposed a significant shift: capturing images of these time series directly from screens. This involved training neural networks to identify patterns not in physical objects but in lines and shapes representing financial data.

Dubbed 'Mondrian', this project extended beyond mere image analysis to delve into understanding trader behaviour through eye gaze tracking. By examining where traders looked most frequently and the sequences they followed, Veloso's team sought to replicate human decision-making processes in AI.

This novel approach was immensely successful. After only a week of training, the neural network achieved an impressive 95% accuracy rate in making buy or sell decisions based on the time series images.

Conclusion

JPMorgan's venture into employing AI for image recognition in trading signifies more than innovation; it demonstrates the immense potential of creative thinking. Under Veloso's stewardship, the company has not only refined its trading strategies but also pioneered new avenues for AI's application in finance and beyond. As AI continues to progress, initiatives like Mondrian will lay the groundwork for future advancements.