When I saw that the head of research at Google, an artificial intelligence researcher from Facebook, the lead machine engineer at Meetup, the CTO of Kaggle, and an eBay data scientist were among the talented people speaking at Machine Learning Conference 2017 (MLC), I signed up immediately. I was so afraid of missing out that I bought my ticket first at $217.27, and only thought to ask about special student pricing while looking at my receipt. My final total came out to $147.77.
When I got there, I could see why I had paid so much money; the whole event was in 230 Fifth Ave. Rooftop with breakfast and lunch included. One piece of advice: If it’s your first time going to a conference like this, I highly recommend getting a lot of sleep and not going with the flu like I did because not everything will interest you. The day was packed with very dense information and if you are new to data science like myself, it will require a lot of mental stamina.
I spoke to quite a few of the businesses sponsoring the event and quite a few were really eager to speak with me. My first stop was SigOpt, where I spoke with CEO, Scott Clark. He described their product as, “an optimization service that uses machine learning parameters and hyper-parameters to try to force it to grid solve”. With more research, I understood that they have a specific method of training data in a feedback loop that gives a result; you adjust the model, and retrain data until you have maximized your output. Essentially, “it allows data scientists to build better models with less trial and error“.
Another interesting conversation that I had was with a recruiter from eBay. EBay, being another big name in the data science world because of the massive amounts of information they work with, is hiring research engineers. Mar said, “eBay is looking for people who are smart, innovative, and driven.” When I asked him what level of experience hired applicants had, he offered, “most come with a lot of experience”. At that point, I felt that it was time to hear some of the panelists. While his answer was fairly vague, it is worth noting that the qualifications for being a data scientist in general are fuzzy. If you can combine your skill-set into a spectacular portfolio of case studies, then you will probably get a job.
I had two favorite talks. The first was a talk from Ross Goodwin who focused on the “narrated reality”. He passionately described how he used machine learning to create art. He showed us words and had us guess if they were real or computer-generated gibberish and proved that it was pretty hard to tell if it was computer generated gibberish. He used this to create a Twitter bot that people could tweet a word at and they received a response with a gibberish meaning. Eventually, Twitter removed his bot. “To me, it felt magical and it felt like Twitter killed my pet,” Goodwin said.
My second favorite talk was by Claudia Perlich. When talking about human predictability, she said, “in two weeks in 2012, across hundreds of different models, we saw a doubling of the median lift and, no, it does not mean the end of free will.” The median lift she is talking about is the predictability of humans and how well data scientists can guess your traits based off of your decisions or, in this case, your clicks. Her conclusion was that bots got smart. These bots are what help websites that pay for their services to achieve a certain ranking by browsing as if they were real-life users. Prediction models were not only working with human behavior, but bot behavior and this cluttered our browsing data. She wrapped up by saying that the people creating algorithms have the ethical responsibility to consider when predictability fails because these models are so engrained in our lives. Some examples she gave were job success based on your LinkedIn profile, where policing resources get allocated, and what ads would be most likely to show up next to your name.
Unlike other MLC headliners, Goodwin and Perlich were amazing not because they were trying to sell a product or convince us that their product was the best, but because they introduced ideas that I felt challenged the audience to think of data in a different way.
Don’t get me wrong though, many of the other panelists had very exciting and important things to say and their presenters were often fairly good, but the event felt too sponsor focused. The sponsors bought the coffee, they spoke about their company, they promoted themselves, and tried to sell us product. If it’s your first time going to a conference like this, I highly recommend getting a lot of sleep and not going with the flu like I did because not everything will interest you. The day was packed with very dense information and if you are new to data science like myself, it will require a lot of mental stamina.
As someone who is not so involved in the industry yet, I would reconsider spending so much money on one event.
If it’s your first time going to a conference like this, I highly recommend getting a lot of sleep and not going with the flu like I did because not everything will interest you. The day was packed with very dense information and if you are new to data science like myself, it will require a lot of mental stamina.
