For the penultimate talk of Day 1, Analytics India Magazine is pleased to have Adele Cutler at the second edition of Rising conference. Dr Cutler is one of the key people behind the Random Forest algorithm. She along with her doctoral advisor, Leo Breiman worked on the development of random forests three decades ago. She and Breiman released the original code for Random Forests and also created the first interface to R. Other works include Archetypal Analysis (1994).
The talk began with an introduction of her co-inventor Leo Breiman and the timeline of Random Forests.
Dr Cutler introduced the virtual audience to the intuition behind Random Forest and how it came into being. She demonstrated this by taking the examples of classification and regression trees(CART) and then went on to explain their disadvantages and how Random Forests come to the rescue.
She explained the advantages of Random Forests over CART, in terms of accuracy and instability. Even if there is a little change in the data, the individual trees will change but the forest is more stable because it is a combination of many trees.
In case of accuracy, Random Forests is competitive with the best known machine learning methods.
She also spoke about the tradeoffs of using Random Forest over the latest approaches such as Deep Learning. She admitted that Random Forests might not be as accurate as deep learning methods for tasks such as image recognition. However, in many other tasks, the improvements offered by current state of the art algorithms is only infinitesimal when compared to Random Forests.
Sharing few experiences from her career that spanned over three decades, she spoke about the involvement of women in algorithmic development. She said that the numbers are underwhelming when compared to statistics, which features more women. Dr Cutler’s talk was followed by a deluge of good questions by the attendees. She tried her best to answer all within the limited time.