Jerry has extensive experience in machine learning, storage systems, online service, distributed systems, virtualization, and OS kernel. He has worked on high performance and large-scale systems at companies such as: Lyft, Box, Twitter, Zynga, and Microsoft. He has also authored the open-source project Lib Crunch and is a three-time Microsoft Gold Star Award winner. Jerry completed his master’s degree in computer science at Shanghai University. His most recent startup is Datatron.
Datatron began in 2016 after you left Lyft. How did you initially conceive of the Datatron business concept?
When we worked at Lyft, we noticed that data scientist usually comes from diverse background like Math, Physics, Bio-Engineering etc. It is often very hard for them to get the engineering part of how their models work although they have good intuition on the model and math. That motivated us to start Datatron. We are not trying to help data scientist to find the best algorithm. We only come into picture after the algorithm is decided and to make the model deployment, monitoring and management more efficient.
Datatron was selected by 500 Startups to be included in the 18th cohort of accelerator companies. How did this residency personally influence you, and how you manage Datatron?
We did learn a lot from StartX and 500 Startup experiences which includes how to pitch to investors, how to find product/market fit, how to run sales/marketing which we don’t have experience personally before.
Datatron is a management platform for ML, AI, and Data Science models. Could you elaborate on some of the functionalities that are offered by your platform?
Our product has four modules now, Model Deployment, Model Minoring, Model Challenger and Model Governance.
Create and scale model deployments in just a few clicks. Deploy models developed in any framework or language.
Make better business decisions to save your team time and money. Monitor model performance and detect model decay as it happens.
Spend less time on model validation, bias detection, and internal audit processes. Go from model development to internal auditing to production faster than ever.
One of the use cases of Datatron is Demand Forecasting which is important for enterprises which need to plan and allocate resources. How does machine learning play into this?
Demand usually changes with both seasonality and trend, which is a typical machine learning problem. Machine learning models like ARIMA, Recurrent Neural Network (RNN) can learn from historic data to find the trend and seasonality automatically and make predictions based on that.
Which framework models (for example, TensorFlow) do you currently support?
We support most of the popular machine learning frameworks like sklearn, TensorFlow, H2O, R, SAS etc.
Which languages do models need to be built in to be supported by Datatron?
We support models in their native languages – Python, R, Java etc.
What are some of the types of industries which are best served by using the Datatron platform?
Fundamentally our platform is a horizontal solution which means it can be used by lots of different industries. As of now, we are trying to focus on financial service and telecommunication.
What are some of the most challenging aspects of data science that companies face, and why does Datatron solve this for them?
Lots of companies have different data science team already and those teams are using different tools to build their model and have different practice to manage their models. More and more enterprise realized that model is becoming an asset and will impact their top line directly. Having a platform can standardize the machine learning practice across the company becomes critical and required. Our platform can help to solve those issues.
Is there anything else that you would like to share about Datatron?
We got lots of inbound interests from big enterprises. At the same time, we are also building up our sales and marketing team to reach out to potential customers actively.
To learn more visit Datatron.
Computer Scientists Tackle Bias in AI
Computer scientists from Princeton and Stanford University are now addressing problems of bias in artificial intelligence (AI). They are working on methods that result in fairer data sets containing images of people. The researchers work closely with ImageNet, which is a database of more than 13 million images. Throughout the past decade, ImageNet has helped advance computer vision. With the use of their methods, the researchers then recommended improvements for the database.
ImageNet includes images of objects, landscapes, and people. Researchers that create machine learning algorithms that classify images use ImageNet as a source of data. Because of the database’s massive size, it was necessary for there to be automated image collection and crowdsourced image annotation. Now, the ImageNet team works to correct biases and other issues. The images often contain people that are unintended consequences of ImageNet’s construction.
Olga Russakovsky is the co-author and an assistant professor of computer science at Princeton.
“Computer vision now works really well, which means it’s being deployed all over the place in all kinds of contexts,” he said. “This means that now is the time for talking about what kind of impact it’s having on the world and thinking about these kinds of fairness issues.”
In the new paper, the ImageNet team systematically identified non-visual concepts and offensive categories. These categories included racial and sexual characterizations, and the team proposed removing them from the database. The team has also developed a tool that allows users to specify and retrieve image sets of people, and it can do so by age, gender expression, and skin color. The goal is to create algorithms that more fairly classify people’s faces and activities in images.
The work done by the researchers was presented on Jan. 30 at the Association for Computing Machinery’s Conference on Fairness, Accountability, and Transparency in Barcelona, Spain.
“There is very much a need for researchers and labs with core technical expertise in this to engage in these kinds of conversations,” said Russakovsky. “Given the reality that we need to collect the data at scale, given the reality that it’s going to be done with crowdsourcing because that’s the most efficient and well-established pipeline, how do we do that in a way that’s fairer — that doesn’t fall into these kinds of prior pitfalls? The core message of this paper is around constructive solutions.”
ImageNet was launched in 2009 by a group of computer scientists at Princeton and Stanford. It was meant to serve as a resource for academic researchers and educators. The creation of the system was led by Princeton alumni and faculty member Fei-Fei Li.
ImageNet was able to become such a large database of labeled images through to the use of crowdsourcing. One of the main platforms used was the Amazon Mechanical Turk (MTurk), and workers were paid to verify candidate images. This caused some problems, and there were many biases and inappropriate categorizations.
Lead author Kaiyu Yang is a graduate student in computer science.
“When you ask people to verify images by selecting the correct ones from a large set of candidates, people feel pressured to select some images and those images tend to be the ones with distinctive or stereotypical features,” he said.
The first part of the study involved filtering out potentially offensive or sensitive person categories from ImageNet. Offensive categories were defined as those that contained profanity or racial or gender slurs. One such sensitive category was the classification of people based on sexual orientation or religion. Twelve graduate students from diverse backgrounds were brought in to annotate the categories, and they were instructed to label a category sensitive if they were unsure of it. About 54% of the categories were eliminated, or 1,593 out of the 2,932 person categories in ImageNet.
MTurk workers then rated the “imageability” of the remaining categories on a scale of 1 to 5. 158 categories were classified as both safe and imageable, rating 4 or higher. These filtered set of categories included more than 133,000 images, which can be highly useful for training computer vision algorithms.
The researchers studied the demographic representation of people in the images, and the level of bias in ImageNet was assessed. Sourced content from search engines often provide results that overrepresent males, light-skinned people, and adults between the ages of 18 and 40.
“People have found that the distributions of demographics in image search results are highly biased, and this is why the distribution in ImageNet is also biased,” said Yang. “In this paper we tried to understand how biased it is, and also to propose a method to balance the distribution.”
The researchers considered three attributes that are also protected under U.S. anti-discrimination laws: skin color, gender expression, and age. The MTurk workers then annotated each attribute of each person in an image.
The results showed that ImageNet’s content has a considerable bias. The most underrepresented were dark-skinned, females, and adults over the age of 40.
A web-interface tool was designed that allows users to obtain a set of images that are demographically balanced in a way that the user chooses.
“We do not want to say what is the correct way to balance the demographics, because it’s not a very straightforward issue,” said Yang. “The distribution could be different in different parts of the world — the distribution of skin colors in the U.S. is different than in countries in Asia, for example. So we leave that question to our user, and we just provide a tool to retrieve a balanced subset of the images.”
The ImageNet team is now working on technical updates to its hardware and database. They are also trying to implement the filtering of the person categories and the rebalancing tool developed in this research. ImageNet is set to be re-released with the updates, along with a call for feedback from the computer vision research community.
The paper was also co-authored by Princeton Ph.D. student Klint Qinami and Assistant Professor of Computer Science Jia Deng. The research was supported by the National Science Foundation.
Data Science Companies Use AI To Protect Environment And Fight Climate Change
As the nations of Earth attempt to invent and implement solutions to the growing threat of climate change, just about every option is on the table. Investing in renewable sources of energy and dropping emissions around the globe are the dominant strategies, but utilizing artificial intelligence can help reduce the damage done by climate change. As reported by Live Mint, artificial intelligence algorithms can help conservationists limit deforestation, protect vulnerable species of animals from climate change, fight poaching, and monitor air pollution.
The data science company Gramener has employed machine learning to help get estimates of the number of penguin colonies across Antarctica by analyzing images taken by camera traps. The size of penguin colonies in Antarctica has decreased dramatically over the course of the past decade, impacted by climate change. In order to help conservation groups and scientists analyze image data of Antarctic penguins, Gramener employed convolutional neural networks to clean up the data, and once the data was clean it was deployed through Microsoft’s data science virtual machine. The model developed by Gramener makes use of penguin density in the captured images in order to achieve estimates of penguin populations faster and more reliably. Gramener also used similar techniques to estimate salmon populations in various rivers.
As LiveMint reported, there are other animal conservation projects that make use of AI as well, such as the Elephant Listening Project designed by Conservation Metrics. Populations of elephants throughout Africa have been suffering because of illegal poaching. The project utilizes machine learning algorithms to identify the vocalizations of elephants, distinguishing them from sounds made by other animals. By training machine learning models to recognize unique sound patterns and then using data from sensors distributed throughout elephant habitat, the researchers can develop a system that alerts them to potential poaching or deforestation. They can have a system that listens for things like vehicles, sounds, or guns, and if these sounds are detected alerts can be sent out to authorities.
Machine learning algorithms can also be used to predict the damage that can be done by severe weather events like thunderstorms and tropical cyclones. For instance, IBM has produced a new high-resolution atmospheric forecasting model intended to track potentially damaging weather events.
Jaspreet Bindra, the author of The Tech Whisperer and expert on digital transformations explained to LiveMint that machine learning is necessary to keep up with the changes caused by climate change. Bindra explained:
“Global warming has changed the way climate modeling is done. Using AI/ML is very important as it will make things happen faster. All this will require lots of computing power and, going forward, quantum computers might play an important role.”
Blue Sky Analytics, based in Gurugram, India, is another example of using machine learning algorithms to protect the environment. An application developed by Blue Sky Analytics is used to monitor industrial emissions and air quality in general. Data is gathered and analyzed through satellite data and sensors at ground level.
It requires a substantial amount of computer power in order to analyze and understand the environmental effects of issues like climate change, poaching, pollution. UC Berkeley is trying to speed up research by crowdsourcing the computation of environmental data using smartphones and PCs. The crowdsourcing project is called BOINC (Berkley Open Infrastructure for Network Computing). Those who want to assist in the crowdsourced data analysis just have to install the BOINC software on a chosen device, and when that device isn’t being used the CPU and GPU resources available will be leveraged to carry out computations.
Garth Rose, CEO of GenRocket, Inc – Interview Series
Garth is the Co-Founder & CEO of GenRocket. He is an expert at launching and building technology startups. He has held numerous senior leadership roles in startups over the past 25 years including President & CEO of Concentric Visions (VC backed + acquired), VP Sales & VP Business Development at Indus River Networks (VC backed + acquired), VP Sales & Marketing at Digital Products (acquired) and National Sales Manager at Leading Edge Products.
In 2012 you Co-founded GenRocket a company that specializes in enterprise test data automation. What was the initial vision that inspired this?
I met GenRocket Co-Founder Hycel Taylor in 2011 and he educated me about the need for accurate, conditioned test data for effective software testing. Hycel had done a lot of research and found a huge gap when it came to test data solutions. Hycel decided to architect his own platform that was low cost, really fast and flexible.
What are some of the benefits of using test data versus production data?
Proper software testing means not just testing “positive” conditions of an application but also testing “negative” conditions as well as permutations and edge cases. Production data is useful for data analytics but has limitations for many test cases. One of our financial services customers shared that their production data can only fully satisfy 33% of their testing requirements.
The speed of data generation is important, what are the speeds that GenRocket can deliver?
For a typical automated test case we deliver test data in about 100 milliseconds. For volume data GenRocket generates at a rate of about 10,000 rows of data per second. For big data applications we can use multiple GenRocket instances in parallel to generate millions to billions of rows of data in minutes.
There’s always a learning curve when it comes to generating both test and production data. Do you offer any type of user training?
GenRocket University was created in 2017 to educate our customers and channel partners on GenRocket. We offer multiple on-line training courses at no cost including our “GenRocket Certified Engineer” training course.
You currently serve enterprise customers in over 10 verticals. What are these different types of enterprise customers?
Major banks, numerous global financial services companies, major U.S. healthcare providers, major manufacturers, global supply chain firms, data information services firms are some of our customers across the world.
Our most active industry verticals are banking, financial services, insurance, healthcare and manufacturing.
How does GenRocket differ from other Test Data Management tools?
Traditional Test Data Management (TDM) solutions copy, mask and refresh production data. These solutions tend to be expensive and complex and production data also has limitations for software testing. GenRocket flips the TDM paradigm by quickly and accurately generating most of the required data and querying the small amount of production data that is needed for some of the tests. The GenRocket Test Data Automation (TDA) approach is faster, lower cost and easier to implement and use than TDM.
Could you tell us a little bit about the ability for test data framework compatibility?
Every organization has their own testing framework or testing tools so GenRocket has the flexibility to integrate into every customer’s environment. GenRocket can integrate with just about any testing framework in any language and any testing tool like Jenkins or Selenium. GenRocket can also insert data into any database, and can send data over web services. GenRocket also offers integration with Salesforce and can support complex data feeds like NACHA in banking and EDI and HL7 for the health care industry.
Is there anything else that you would like to tell us about GenRocket?
We rely on an extensive network of trained channel partners to introduce and deliver GenRocket test data solutions into our global customers. Partners like Cognizant, HCL, Wipro, Hexaware, Mindtree and UST Global are actively working with GenRocket.
To learn more visit GenRocket.