Finance & Technology

  

Financial Stress Tests – Generative Deep Learning


Post 2008 Financial Crisis, regulations such as Dodd Frank and CCAR have tried to reign in risks taken by banks. But regulators face two challenges from big banks - (1) balance sheets that are too complex to decipher and (2) lobbying due to any perceived inordinate threat to profitability. So regulators must devise the smallest intervention with greatest rewards. At the very outset we must realize that any Artificial Intelligence (AI) solution may face a faster, better, adversarial AI built by the banks. We can not let the chips fall where they may.


Read More ...



Bitcoin Disruption in Payments – Winners and Losers


Satoshi Nakamoto noted that traditional payment networks allow reversible transactions but gain centralized intermediary power in return. We show that Bitcoin does indeed help users that need to irreversibly transact large values. But what about the rest of the users?


Read More ...



Blockchains – For Business Researchers


  

Bitcoin and other crypto currencies have shot to fame over the last couple of years. Academic research on sub fields such as cryptography, security and the technology stack has burgeoned. However economist and business researchers are yet to analyse whether it should be considered a revolution, both from the point of view of collaborative micro-economies these coins create as well as broader impacts on fiat economy. I make a modest attempt at summarizing a few topics that should interest researchers interested in getting started with these questions.


Read More ...


Deep Learning & Economics

  

Deep Learning grounded in Structural Econometrics


A typical Machine Learning algorithm builds a model of the world using only observed outcomes. In contrast, Structural IO allows an economist to enforce their theory. If this theory is “correct” it avoids over fitting to the specific observed sample. This raises a key philosophical question on what should be considered ground truth. Economist often rely on convincing arguments of human behavior prior to exploring how well observed sample fits their prediction. A statistician favors learning from observed sample as their typical task do not need a coherent and rational model of human behavior. Perhaps we can combine strengths of both these concepts.


Read More ...


  

Dynamic Structural Econometrics and Reinforcement Learning


Social Scientist often deal with multiple humans simultaneously attempting to understand their environment and achieve their goals. A bunch of teenagers learning the rules and tricks of poker while also trying to win every hand. Ketchup producers trying to comprehend consumers & competitors while trying to make profits. Economist seek to observe such interactions in order to infer different motives (utility functions) driving these participants.


In Computer Science, Reinforcement Learning (RL) algorithms similarly designs robots to explore the world around them and exploit it for their goals. Unlike humans, we can inspect the precise logic prompting a robots every single action. Since these robots have started to show human like sophistication could we use them to better speculate human behavior. 


Read More ...

Visual Perception & Economics

Beauty Bias


Alongside gender and racial biases, recent popular media has highlighted the impact of sexual harassment at workplace. While the limelight is new, most of us have always pondered on these questions and perhaps even adjusted our behavior in social situation to adhere to these unspoken stereotypes and perceptions. Our research studies career trajectory of university graduates from last 30 years to unearth biases suffered by individuals with different looks who otherwise look comparable on paper. We train an AI (Artificial Intelligence) agent to mimic human perception from an individuals looks; characteristics such as attractiveness, race and personality. We specifically answer questions on bias - 1) how much($ salary lost in entire career), 2) what career phase (graduate school vs 1st job vs mid-career vs late career), 3) which industry (tech vs finance vs marketing), 4) what jobs (software dev vs sales vs accountant) 5) differences between genders.


Read More ...


Advertisement Perceptions


An increasing portion of retail shopping happens online. Purchasing computers, phones and groceries online is natural, as these (“search”) goods are standardised and one can rely on their own choice of specifications among various options. More surprisingly people have quickly become comfortable buying (“experience”) products such as clothing, footwear and other accessories online. Apparel brands must therefore compete side by side for customers attention just relying on fashion photography visuals. Customer who often navigate 1000’s of product options must swipe away most options and narrow down quickly to save time (Obviously some of us are happy spending more time than others at shopping!). Our research studies what photographs are best – 1) photographs background 2) Age, ethnicity, looks and expressions of the fashion model, 3) Model pose (confident, gentle, threatening). Most interestingly how characteristics such as haughty expressions may work for a high end expensive product but drives people away if a cheap brand attempts the same. 


(In collaboration with Adobe)


Virtual Reality (almost!)


Virtual Reality and Augmented Reality have captured our imagination and its been touted as the next big breakthrough in the technology market. Our VR-like setup back in 2010 allowed a user to interact with a three dimensional remote scene in real time by physical motion as well as nuanced head movements. Unfortunately we never quite saw the potential nor did we actually call it VR at the time. We mostly focussed on the limited application at remote (deep water or planetary) interactive exploration. We solved problems such real time scene rendering with GPU’s, integrating a 3D television, human pose & head tracking and all with a single camera to capture the scene