About the talk
Colman Madden (Customer Engineer) and David Li (Director of Advanced Analytics at BlackRock) present on how BlackRock leverages NLP on Google Cloud to search for signals in the investment management industry.
Learn more about topics varying from collecting and manipulating data on Dataproc to extracting features via BigQuery, and finally estimating topic distributions for signal generation.
Speakers: David Li, Colman Madden
Google Cloud Next ’20: OnAir → https://goo.gle/next2020
Subscribe to the GCP Channel → https://goo.gle/GCP
product: Cloud Document AI API, TensorFlow Enterprise Core, AI Platform; fullname: Colman Madden;
event: Google Cloud Next 2020; re_ty: Publish;
I don't want. My name is Coleman. Focus on Asset Management, hedge fund industry in space. I want to hear what they believe from BlackRock. I'm up in collaborating with since early. 2019, David over to you. I work in the financial modeling group at black rock group of Engineers and researchers working together. What percentage of how to construct a dictionary? Foundation. we will deliberate manner Services provided We have a wide range of skills and solutions to see my teacher in Erie tomorrow.
Information extraction. The Fresh Market. Oh, so you to find the nature of these Investments. Level Rod Morningside model, which is he. Where is the year u.s. Rivers loans Market was around 1.2 trillion dollars in 2018 and 2019, a couple was company to invest in the market. Are becoming increasingly difficult, for example, of the political landscape, and social and governance. Some of the challenges in predicting issue or CD. Listen to nobody has actually available
available news for flexible where they typically are constrained in a fixed competition plan. Will it rain in Chester ideas so we can extract and classify text. Asian salmon and cream and then Add clusters with Sea Cloud in Cloud console. Recent message from estate. After that let me speak to reach 100 and expecting filter relevant information for me. There's no shooting or configuration involved in or magically handle the workload and produces results quickly. Will you say I know book and cloud storage of the combination of our
training and evaluation cycle up and down? Usually, operating on premise, we need to be very careful about resource consumption about an hour. Find the scalability and flexibility to change the clutches of the type of cost too much computer. Some weapons. We learned during this exercise. Securus or second pumping station for the car. Ships between entities. Examples of applications are the company's revenues and expenses. Samsung is a supplier of screen for Apple there for a Performance Machine Company.
Corporate structure. Call Performance of one company. Luckily, we were talking about yesterday before reading models in measuring corporate. Governance was calculating the diversity of experience. And expertise of company performance. Right hand side of a partial Melissa, it shows the ownership relations of like one of those Google and the two companies are connected to all of that which is the parent company of Google. Give me a sentence. End. The production process of these model and we show how we can use Google Cloud.
Wake me up at 4 a.m. architecture. We can see where the remote interpreter on the virtual machines to on computer engine, different types of lesion on a local machine. Would you put our model model to find some Nati? Through Powhatan Point to Great use cases, where is T-Mobile in leveraging gcp, at Black Rock and I want us talk a little bit more. Broadly on how the investment industry is partnering with Google's bad. I don't think it's any secret that the buy side has become increasingly dependent on did any light upon investment
decisions and you'll see that by looking at some of the statistics on the right-hand side of the slide, Google. We believe this lines really well with our powerful analytics platform and they are research coming out of Google brain which we provide for a customer's Choice outside form. when we stopped by looking at a typical investment decision flow 4 by 4, Berkeley formulate, an investment idea Source data crew out that a hypothesis generator signals from the data. Instructables portfolio Baptist. The portfolio presented findings in the vacation from the investment
committee. The blue box highlight, the stage is a process, we kill are a great fit for Google stata. The weather in investment research on Google Cloud to fly the ability from basking. Team, spin a project quickly or react to event in the market and passion to streamline all morning and new data sets from commercial data vendors, such as refinitiv, and see me group, Troy did a Marketplace three first computer clothes into the cloud and limited resource contention for Mom friend computer. Or get access to the latest, AR researcher, apis amount
of services and five orchestrate around my work clothes and MLS platform. So, how do we do all this on Google Play. Is the height of a large picture of how friends are running investment, research on CP on board. As such as Market, portfolio news data sets, with yellow, boxes below. This is where the research teams are using their tools access to data sets above. Understand is often a wall between different basketball teams and we can implement this user access management. Tools while still allowing access to Shared resources, such
as tickets 3 left side into the GCB products, the sibylline these boxes, And it did our position project. We can adjust streaming dead. Derribar pops, a message to you or landfalls on Google Cloud Storage jobs in Proctorville, Freakshow. We're all sitting a lot of customers moving to an elt architecture with a loaded in as quickly as possible. And then use the power of inquiry to actually do that transformation. You can see people Delta Marketplace allows better Publishers. I can get hers. Sure, the data sets with
financial firms without the need for any tail lights. Give me a sandwich just now to pick Corey almond, which will enable firms to create the bid on AWS or Azure single interface without any process of moving up our copies of data. We push the compute to the data The research projects wanted and researchers can run simulations on his PC. Clusters 100,000 kohrs, Kustoms, yams, a perfect. The ratio of course, memory options. A nice run for less than 24 hours at
80%, discount and memory. I'm going to ride our sorry. I wanted and researchers to create a line of books with the latest tensorflow pytorch. And so I can learn images on GPS or cheap used on pull this data, to straight into pandas dataframe. Auto email login to Craig, machine learning models for natural language structured data translation. Without how to write any code is miles, can be trained to understand terminology that is specific to the financial services industry. Are dolphin. In the office Solutions, need to process complex Financial
filings and earnings reports with cutting-edge. And I'll be in trust for learning Technologies. Phone tf-x, Ops solution to evaluate rain in the Playboy machine learning models production, One really important thing to note here is that this is all deployed by m structure is code. Using something like the plant manager or terraform Scripts. So how'd you get started? So step one. You want to get the cloud foundations and place understand your requirements around networking security. Access management wrap these up and infrastructure as code. MTV's into a project
creation part 1. Secondly, you want to understand the process for on boarding, new data sets and pull the data acquisition pipeline to automate this thoroughly understand what tools do researchers need an automated deployment of tools along with the connectivity. They need access to data required, to run their experiments to make life easier for everyone. Thank you for coming along for the talk today, please. Any questions to the Dory or feel free to reach out to me or David directly?
Buy this talk
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.