Syed and I changed venues to El Jannah in Granville for today's meeting. Ever since Elliott took me to this place two weeks ago, I have been looking for an excuse to come back.
"It is from Arabic. In our religion, when you die, you are judged. If you are good, you go to 'aljana', if you are bad, go to 'aljahim'. El Jannah is from aljana."
"So, we were just in heaven."
We are going to get fat in heaven.
Today's meeting lasted a semi-marathon three and a half hours. Earlier this week, when I was at the Zambesi launch, when I was supposed to be listening to the talks and mingling with the crowd of startups, I was actually writing the latest iteration of our recommender system. I had been working on a way to simplify the probability density functions in our propensity algorithms into simple operations that can be done on any mobile phone in real time.
I think I figured out a way to achieve approximately the same mathematical outcome using three layers of simple arithmetic operations - turning this monster truck into a Ferrari. Unfortunately, I cannot share the solution on the blog! If what I propose works, then we might have cracked something pretty amazing, increasing the power of our algorithms and increasing the speed simultaneously, and that would be a proprietary technology.
The new tech is probably six months away from implementation. We will have to write the codes and test them first.
Ian and I had a 3 hour coffee this morning to talk about VIVA's data structure and algorithms. Like with Tom yesterday, I forgot to take a photo of Ian.
I keep forgetting to create content for this blog! This is why I have taken a crappy photo of tonight's launch of Zambesi from the back of the room.
This subpar photo was taken before approximately 25% of the audience slowly trickled away.
Going to the Zambesi launch tonight reminded me how powerful it will be when we refine VIVA's current algorithms to read beneath the superficial self-reported layer of information - figure out what an event *actually* is, and then show that event to the people who *actually* want to be at that event.
I managed to squeeze in a meeting with the head of Sydney's machine learning meetup group, Paul Conyngham, before the Zambesi launch.
"So it is a feed-forward neural network? Do you know that term?"
"Yep. I know it."
"You should use the right terms. You would sound smarter."
Oh, that was blunt, I thought...
I have never referred to the VIVA algorithms as that before. I didn't even know I was creating a machine learning algorithm when I first designed VIVA... lucky for me, Steve is around to give me some vocabulary in the area... though apparently not enough.
It has been a busy couple of days at VIVA central, and so the blogging has fallen by the wayside a little - as I was saying a week ago, I wish I had a clone... though I doubt my clone could have done any of the coding that Syed was doing with any proficiency... so I think I just want a cloning machine and then make clones of people as and when required.
I haven't left my home office for days. People tell me that the weather in Sydney over the weekend was amazing.
"You don't say," I would answer. I had no idea.
Reece and I caught up this morning about a starting new project we are calling Project Klink. It uses the same sorts of machine learning algorithms as VIVA to solve a very different problem. I am optimistic about this project because it is another means of validating VIVA's algorithms, and I expect I will learn a lot from Project Klink to help make VIVA even better. We haven't formed plans yet, but they will surely come in time.
I had a coffee with Tom as well this morning. Tom used to help us with the business development side of VIVA. He was the one who connected us with the marketing manger at Bell Shakespeare. It has been about two months since we chatted - entirely my fault, for being absorbed in my own narrow world.
We swapped our latest stories. I told Tom about the happenings with VIVA, and the latest news of how we received a nice rejection email from TechCrunch this morning:
I wondered if machine learning is not sexy enough.
Tom told me a little story about how far VIVA is spreading, "A mate of mine asked me about VIVA the other day. His mates told him about VIVA, and asked if that was the app I was working on. Yeah it was! I had been telling him for months, and then he heard about it from a mate of a mate!"
I messaged the team on Slack something to the effect of, "Team, it is a pity about TechCrunch. I guess we will just have to make VIVA a success the old fashioned way!"
Elliott, Ruth and I went to El Jannah in Granville for dinner last Saturday night (so this blog post is a #latergram of sorts - which is unprecedent on this blog - but since I have been stuck inside working on this beautiful summer's weekend and I have no other photo to share, this will have to do).
There is a map about the socioeconomic divides in Sydney illustrated by chicken shops... Ruth showed it to me when we were walking to the car. That is a blog piece for another time.
The reason I wanted to put up a blog piece tonight is because I stumbled across what is possibly the clearest explanation of the functions of neurons and layers in a neural network that I have ever seen - and it is deserving of joyous celebrations:
I went to my first Sydney Machine Learning Meetup today. Anthony at The Last Pickle recommended joining this group when we met about a month ago, and I happened to stumble across one of their ads on Facebook or something late September. I invited Ian and Syed to join me tonight, but tickets were sold out by the time they looked, so I went solo (I read a great joke about Han Solo once upon a time, but I cannot remember it right now).
It was a surprisingly good event. Now I want to learn more of the development side for implementing machine learning so I can do it myself - instead of relying on others to implement my formulas.
There is a lot of buzz around using TensorFlow, and one of the speakers tonight is using it in Pascal 51. Frankly, I don't know enough about TensorFlow to have an opinion on the matter. My friend Jon in New York knows a lot about it, but he is in New York and I cannot just knock on his door for a beer and a chat. Steve might know something about TensorFlow as well, but he lives in Berlin and already has a heap on his plate.
I still want to learn though, especially after tonight. Maybe Nishant and I can have a go at it together. He already knows how to code in Python, so he is a step ahead of me.
Damn, I need to get their events onto VIVA.
Data Skeptic has become one of my favourite podcasts.
I refused to listen to podcasts until a few months ago, but the combined influences of Reece at WWF and Tina at Sydney Festival convinced me to give them a chance.
My prejudice wasn't against podcasts specifically, but against anything purely aural. I think it is because I cannot see what I am getting into until after committing the time to listen - there is no window shopping, only unknowing commitment and the risk of disappointment.
Some people love the exploration and possibility of discovering something delightful in the soundscape of the world. I don't know why I have never been one of them... I love new experiences. Maybe it is a time thing. Maybe it is an allergic reaction to the risk of disappointment.
Yesterday, I found Kyle Polich at Data Skeptic had released a new podcast about recommender systems. This is a significant thing because VIVA is powered by our proprietary recommender system, and because I do not have a long history in the machine learning industry, every bit of information about what is at the cutting edge of this machine learning technology matters to the fate of VIVA. Well, I am happy to report that VIVA still seems to be at the cutting edge.
The news warranted a break, so Allan and I went to The Buena for a trivia night.
(despite the consistency of this week and last week, I have only been to trivia twice in the last few years - this week's trivia was much easier than the last)