Research

We periodically publish our own independent research on various AI and related topics. Unless otherwise noted, all opinions are those of the authors and do not represent those of Innospark Ventures.

AI & Learning: A Preferred Future
Dr. Venkat Srinivasan, March 2022

Rapid progress in AI technologies has generated considerable interest in their potential to address challenges in every field and education is no exception. Improving learning outcomes and providing relevant education to all have been dominant themes universally, both in the developed and developing world. And they have taken on greater significance in the current era of technology-driven personalization.

Learning outcomes have however stagnated in the U.S. and have remained relatively low in the developing world. In their 2020 annual letter, Bill and Melinda Gates point out that despite spending billions they have not seen the kind of progress they expected with learning outcomes in k-12 education in public schools (CNBC, 2020). In India, each year, the Annual Status of Education report reports in India’s public schools, a 5th grader is unable to read 2nd grade text (ASER, 2019). Recently, the Central Square Foundation (Central Square Foundation, 2020) found that learning outcomes in private schools in India, while better compared to public schools, are still poor – 35% of rural private school students in Grade 5 cannot read a grade 2 level paragraph.

Read full article

Towards a New Architecture for Artificial Intelligence

Dr. Venkat Srinivasan, August 2018

AI is in a renaissance powered by the availability of enormous amounts of data, connectedness and the low cost of computing infrastructure. Yet, we believe AI is in its early innings and has to overcome significant challenges in order to deliver on its promise. Indeed, for AI to be successful, we believe the early exuberance around algorithmic acquisition of intelligence has to evolve to a more robust context aware, traceable architecture for intelligence acquisition which can also function in sparse data environments.

We propose a new architecture for AI grounded in principles of transparency, contextretention and ‘computational abstractions’. Computational abstractions, encapsulating structured representations of knowledge in various domains, both abstract and specific, can enable speedier, more validated knowledge acquisition for AI solutions including in sparse data environments. We hope the architecture contributes to and advances successful AI applications in the real world.

Request a Copy

The Future of Jobs, Skills and Education

Dr. Venkat Srinivasan, July 2018

There is widespread debate on the impact of AI on jobs, skills and education. While estimates of job losses range from small to very large, the consensus is that there will be widespread job dislocation and skill shifts. Corporations, educational institutions and individuals alike are grappling with how they should react and stay ahead of these developments.

Prior research has attempted to illustrate the potential impact of AI at a macro level. In this paper, we attempt to develop a planning framework at the micro level. We illustrate the impact of AI led automation using real world business processes as examples. We analyze the implied skills shift in the pre and post AI world. Based on real world examples and skills analyses, we outline a simple dependency model between occupations, skills and education using the Bureau of Labor Statistics (BLS) definitions. We then overlay a subjective estimate of automatability of the primary tasks of various occupations as defined by BLS. This provides the mechanism to estimate the micro level impact of AI and automation on specific occupations, skills and education programs. The resulting framework can be useful for policy makers, corporations, educational institutions and individuals.

Request a Copy

Machine Learning and Corporate Bankruptcy Risk Prediction

Dr. Venkat Srinivasan, July 2017

This paper presents the latest results of our work on bankruptcy prediction. The use of statistical models, both parametric and non-parametric, for predicting corporate bankruptcy has been the subject of intense scrutiny in academe over the last century.  The economic crisis of 2008 again brought center stage the need to proactively monitor and predict defaults.

In this paper, we present the findings from applying several machine learning methods to predict bankruptcies using fundamental financial metrics from a proprietary database of bankruptcies in the U.S.  A major objective of the study is to assess the comparative efficacy of traceable methods like CART.  Methodologically, the study differs from prior research in two significant ways.  First, we make an attempt to normalize the financial data before training the machine. Second, we adopt a more robust validation approach for the machine learning models compared to prior research.

We provide predictive results up to 2 years prior to the bankruptcy event.  Results show that traceable deep learning methods like CART perform as well as non-traceable methods like Neural Networks.

Request a Copy