6PM - Entry
There will be pizza, courtesy of ComplyAdvantage (https://complyadvantage.com/).
6:30 - Transaction Monitoring and Money Laundering (Cristi Persic, Oana Ratiu, Ariel Pontes)
Every day, millions of transactions between people or organizations take place, amongst which there might, just might, be some that are suspicious. Our job is to identify these transactions and why they do not fit a normal pattern. Hence, why we built our transaction monitoring platform. We will be talking about the reasons why this platform needed to be both generic and flexible and how we achieved this. We will go on to discuss:
• Why different clients have different needs depending on the information they choose to store on their transactions. Why traditional SQL tables are not entirely suited to this application and why NoSQL is a natural alternative, as we still need all the traditional layers of validation and normalization of external input.
• The challenge of building an engine that is optimised to process large numbers of transactions, whilst at the same time being flexible enough to support all the different styles of analysis that are required to detect money laundering and terrorist financing.
• And in terms of performance how far you can take ORM and when it is acceptable to take matters into your own hands. Finally, we will discuss splitting business logic between python code and database queries.
7:15 - Break
7:30 - Machine Learning Supermarket Paralysis (Cristi Lungu)
Over the last three years deep learning has exploded in popularity and the quality of the results has greatly improved. Big tech companies have increased their research budgets in order to benefit from this revolution. As a direct consequence several of these (Microsoft, Google, Facebook, Amazon) have released open frameworks for training and developing deep architectures such as: TensorFlow, CNTK, FBLearner, MxNET and Caffe. Beside these, other popular existing frameworks and wrappers coming mainly from academia, have gathered a great deal of support (Theano, Torch7, Scikit-learn, Keras). With such a broad choice, it's no surprise that choosing the "right" Machine Learning (ML) framework can lead to a paradox of choice (especially for beginner ML students). This talk will give an overview of what ML frameworks are and what they do. We will discuss in more detail TensorFlow, Scikit-learn, Theano and Keras, highlighting the pros and cons for each of these frameworks. Finally, we will demonstrate them by applying them to a simple linear regression task.
8:15 - Closing
Ariel Pontes (Python Engineer) studied Computer Engineering at the Catholic University of Rio de Janeiro. After learning Python/Django in his first internship in a fintech startup he fell in love with MVC frameworks, and re-wrote his final year project in Ruby on Rails. In 2014 he moved to Cluj and started working as a full-stack Django developer in a big outsourcing company. Ariel joined ComplyAdvantage in 2016 where he works on the transaction monitoring and screening platforms. Outside of work you can find him participating in activism for local NGOs and preaching about secular ethics.
Oana Ratiu (Python Engineer) graduated from Babes-Bolyai University, Cluj-Napoca, studying computer science, and is currently completing her Master's Degree in Databases. She has worked as a programmer for three years, mostly with Python. Excited to fight terrorism through her work, Oana joined ComplyAdvantage in early 2016, working on a transaction monitoring platform and, more recently, on a screening platform.
Cristian Lungu (Senior Python/ML Engineer) wrote (anti)viruses for fun and profit for 5 years until he realised how much money flows through the stock market. He then spent his next 5 years coding derivatives, futures and options. Somehow, he was kidnapped and dropped on this startup ship called ComplyAdvantage where fighting terrorists and money laundering with machine learning is the only way to get a meal. As a result of this event, he challenged himself to actually finish his lifelong hobby, a PhD in machine learning, this century. When he doesn't code he's probably teaching bubble sort at the UTCN or conducting practical studies of gradient descent on some mountains.