Below is adopted from forward in my Ph.D disseration.
A man’s character is his fate.
- Heraclitus
It all began with the naıve and lazy man’s dream. I wanted interesting information to come to me even though I didn’t know if such information existed, and I didn’t search for it. I am lazy but addicted to information. I am also a Maximizer. According to the book The Paradox of Choice: Why More Is Less by Barry Schwartz, a Maximizer is the kind of person who scans all the available cereals in the supermarket and tries to select the best one. Frankly speaking, I envy the Satisficers because they will choose whatever option meets the requirements and forget about the rest.
That’s how I initially became interested in recommender systems, which are a class of algorithms that recommend something to my taste, as Amazon and other web-based services that are trying hard to extract more money from us. Also, from my previous experiences with wearable technology, I knew that the ability to extract valuable information from data will be the key component in the so-called big data value chain. Without the ability to turn data into insights, big data is just investment and cost. Analytics will be what generates revenue and profit.
However, the journey never goes as expected and you never know where you will end up when you are setting out. Such was my Ph.D. I started with recommender systems, but the recommendations they make are not satisfactory due to limitations in the quality of these algorithms. It may be rather contentious to suggest that recommender systems algorithms are limited. Indeed, the world is changing and you never know if the next big scientific breakthrough will improve them to be more effective. But as anecdotal evidence to support my claim, Netflix never ended up using the state-of-art algorithms from the famous 1 million dollar competition. The algorithm showed top-performance, but the performance gain was not meaningful to justify the algorithm implementation cost.
That’s when I started to look for alternatives. Did I already tell you that I am a Maximizer? Therefore I concluded that we need to amplify the cognitive ability of the user to tackle the challenges of big data value creation. That was the focus of my Ph.D., presented here, with five design studies. The ideal ending will be that I am satisfied with my methodology and live happily ever after. But after six years of study, I see some fundamental limitations in the visual analytics (VA) approach as well. Those limitations are 1) VA systems are application/domain specific, 2) dependence on back-end algorithms that usually rely on the bag-of-words model, and 3) the requirement of user intervention (this can be both desirable and undesirable. But I am rather lazy.)
So here I am packing light again to find more fundamental ways to achieve my dream. Recent advances in neural networking look promising, and, being inspired by those advances, I want to build upon them to start a new journey into this untapped area. I am excited, afraid, humble, and foolish on this new journey.
If you want to know more detail, you can read my research statement.