It feels like waking up after a long, deep dream. Chronologically short, but memories that stretch afar and run incisively baritone. Rich and sparkling, full of wonderment and magic. A flurry of challenges and maladies that come with them. Anesthetic at times to even sharp pain, becoming a frequent habitat of the bleeding edge. I cannot grasp how they could all fit into just two short years. How such compression is possible I would never know. One would be otherwise incredulous except that it all feels true. The fact that I have finally begun to write again after such a long hiatus suggests that I am well rested. That my decompression is complete. It is so welcome!
When I arrived at MIT almost two years back, l expected to be challenged. But I did not expect to be changed. I can vouch very safely that the past two years have been by far the most breathtaking in the twenty six years that I’ve existed. I’ve learned more in these two years than ever before. Not just about bleeding edge computational science, but of the nature of human beings and of life itself. Glimpses of the magical and the possible. I’ve found that I have more resilience and tolerance in me than I had ever imagined. Most of all, it has come to dawn on me that the friends I’ve made in these two years are going to stay with me for life.
Sparkles in the Brownian movement
The more you learn about the mysterious way the human brain functions, the deeper you probe attributes of aspergers and autism, the richer and more pervasive your insights will be into the world of statistical learning. The world is a well-oiled mixture of distributions. You learn quickly that shallow representations of a phenomenon is not of much use; there is no such thing as a whole. Most phenomena are a complex set of distributions, each talking to one another, each functioning in part on their own and in part because of other parts. Hierarchical by nature. There cannot be a one-size-fits-all mindset that still plagues almost all of the machine learning world. There are lots of lessons to derive from human physiology and from biological systems in general.
Viewed from afar, it might look like uncontrolled Brownian motion. A world of statistical learning that focuses on stable patterns, generalization and abstraction, but the real world which focuses on specificity and uniqueness. How can these two vastly different worlds be bridged? It would seem like finding order in an Brownian motion, a futile and hopeless exercise. But observing how nature has crafted our many functions is a start. Deep learning is the direction to go in. Enough of one-size-fits-all approaches like singular vector machines and conditional random fields. More probabilistic graphical models and deep learning. Careful examination of phenomena. Identifying sub-distributions and tackling them appropriately. A cascading set of processes where sub-distributions talk to each other. A well-oiled machine. There is meaning in the Brownian motion. It sparkles with simplicity.
An assault on cancer
Not too long ago, I came across this fascinating work by Marty Tanenbaum of Stanford. When he was diagnosed with a form of melanoma, he was struck by the different prognoses given by different oncologists. Marty mined the Stanford Cancer Research database which combines knowledge sources from drug trails and a complete medical and treatment histories of thousands of cancer patients visiting Stanford Medical Hospital. Marty was able to give himself a set of treatment options and drug prescriptions by mining this database. He is cancer free today. He speaks of classes of drugs for classes of people for classes of ailments. A probabilistic approach.
One cannot but think of what we can do if every hospital and research medical institutes in the world begin to log the entire medical and treatment histories of their patients. If details of every drug trial conducted anywhere were all logged. If we aggregated research from cancer gene therapy and translational medicine. We may have already discovered a set of cures for certain types and forms of cancer, but they might all be buried in disparate data, strewn away from each other. We have gigantic human endeavors like the Google and Bing search engines, which are essentially mammoth supervised learning agents acting on gargantuan amounts of data. One can only imagine of what we can do if we design something of a similar scale to connect every single resource connected to cancer research from all over the world.
Increment true counters
I have learned many a thing during the last two years. Opportunities have swung by almost thrice as fast as the challenges, as real and painful the challenges can be. Many a time, it is easy to find yourself unable to choose. A paralysis of sorts, the frustrating state of being rendered immobile. I have come to realize that the only things that matter are my friends and my family. In the end, those are the only things that matter. There are people in MIT that have done so much for me. They will do anything for me and I for them. I am very fiercely fond of my friends. I will gladly take the summer heat and the snow monster because it feels so completely right in Boston.
Life is not a never-ending Markov chain. During their journey in the dark, when they pass through the mines of Moria, Frodo laments that he wishes the ring had never come to him, and that he wishes for a lot of things to have never happened to him. So do all who see such times replies Gandalf, that it is not for us to decide; all we can do is to decide what to do in the time that is given to us.
When looked at and listened to carefully, there are sparkles in the Brownian motion.