A few years ago Amazon has startet to replenish its business with Big Data Services – and for that opened its Amazon Development Center for machine learning in Berlin, where authors of e-books meet engineers and engineers meet startups. Dr. Ralf Herbrich, Director of Machine Learning Science, has answered our questions.
[highlight_area classes=“brand“ headline=““ button=“Dieses Interview auf Deutsch lesen“ button_link=“https://bigdatablog.de/2015/10/15/amazon-development-center-fuer-machine-learning/“ button_target=“_blank“ icon=“false“][/highlight_area]
Since 2013 Amazon Web Services is located in Berlin, offering Big Data Services like Sentiment Analysis using data from Twitter. In which way are the services of the new Amazon Development Centre different?
Amazon is constantly innovating on behalf of our customers and the Berlin Development Center is a great example of this. At the Berlin Development Center we are working to develop next-generation technologies for Amazon with a particular focus on the area of machine learning and distributed systems. The Berlin Development Center started in a temporary office in Berlin Charlottenburg from September 2013 until June 2015 when we opened our new Berlin office in Krausenhöfe.
Among other teams, the office is home to the Amazon Web Services (AWS) team, the AWS OpsWorks team as well as the Amazon Machine Learning team. Some of the technologies developed by these teams have already been launched and made available for customers, such as Amazon Machine Learning and AWS OpsWorks. All of these technologies enable both Amazon and AWS to deliver new capabilities to our customers, whether that be new features for AWS or internal capabilities that allow us to get packages in the hands of Amazon retail customers quicker.
[pullquote orientation=“left“]
Dr. Ralf Herbrich, Dircetor of Machine Learning Science, Amazon
[/pullquote]
With the methods of Predictive Analytics it is possible to perceive patterns and to predict future events with a certain probability. Those patterns also allow to forecast trends in fashion. Is it also possible to detect the personal style of the customers and consider it for offerings?
Methods of machine learning or predictive analytics allow detecting patterns in data. Such patterns included dependencies between fields of the data or rules that can reproduce fields of the data with high accuracy, for example the propensity to buy a product depending on colour, texture or brand of a fashion product. Applying methods of predictive analytics to customer-level data allows them to extract personal style which can be used for personalized recommendations. As this is the case we frequently will send style recommendations to customers.
The new location, the Krausenhöfe in Berlin Mitte, is right in the middle of the centre of the publishing houses. Does this have a symbolic meaning – is Amazon focusing again on the publishing business? Or are the services of the Amazon Development Centre for customers of all kinds?
At the Berlin Development Center we are working to develop next-generation technologies for Amazon with a particular focus on the area of machine learning and distributed systems. The machine learning team works with all businesses of Amazon, ranging from retail (sellers, retail customers, fulfilment), Amazon Web Services and digital media and devices – the services that the teams in the Berlin Development Center are working on is for all customers of Amazon.
You also want to work together with authors of e-books, to improve your services for them. How do you organize the collaboration between authors and developers? Aren’t there any communication problems?
To coincide with the Berlin office opening in June this year we held the inaugural Amazon Academy, which brought together authors, seller, developers and scientists for a whole day to exchange their problems and ideas with each other. This was well received by our customers and we will continue this format to facilitate the communication between authors and developers. We are currently planning the next Amazon Academy and will announce date and location in the next few months.
[selectivetweet]The @Amazon Development Center in Berlin: An Interview about the Future of #BigData and #MachineLearning.[/selectivetweet]
One of your core themes is Machine Learning. How exactly is Machine Learning working in general and specifically for text processing?
Machine learning is the science of detecting patterns in data and using them for accuracte predictions and optimal actions – in some regards it is „Statistics of Big Data“. The methods used in machine learning are search methods in a set of rules that could underly data – the accuracy with which the rules explain recorded data is the search criterion. For text data, the rules considered in machine learning methods use words as basic building blocks of the rules and often first try to predict which part of speech each word belongs to (e.g., verb, noun, noun phrase) to then use this prediction to extract more higher-level meaning such as sentiment.
What are the limits of Machine Learning relating to text processing: for example, is it possible to recognize irony in texts, say in those of customers in the feedback section?
Extracting irony in texts is, in principle, possible. However, there are two challenges: firstly, irony is very subjective and a machine learning method is only going to be as good as the quality of the training data it is provided. If two people with a completely different sense of irony provide annotations for a machine learning system, it won’t be able to predict irony well – unless the identity of the labeller is known in which case a labeller-dependent rule can be learned. Secondly, it requires the machine learning methods to comprehend an understanding of the text, in particular which entities are staying in relation to each other. The problem of named-entity and relationship extraction from raw text is at the forefront of machine learning methods for text.
The Amazon Development Centre also aims to fund and support startups. What ways of funding/supporting are there going to be? Is the focus here also on Machine Learning?
AWS has been supporting Berlin startups since we launched the company over nine years ago. Startup companies in Germany looking to get started with cloud computing can get going for free using a programme called AWS Activate. AWS Activate is designed to provide startups with the resources they need to get started in the cloud and gives free access to technical training, technical support and 1:1 help from technical professionals. It also includes the AWS Free Tier, which gives startups access to free technology resources that they can use to get their idea off the ground with no cost to them. For more information on AWS Activate I would recommend visiting our AWS Activate Homepage.
In addition to this we will be opening the AWS Pop-up Loft on the fifth floor of our Berlin office in October. The AWS Pop-up Loft gives startups a space, where they can meet with AWS staff and get 1:1 advice from Solutions Architects, attend technical bootcamps and trainings, run through self-paced, hands-on-labs and attend technical sessions from AWS staff and partners. This will be one of only two AWS Pop-up Loft’s in Europe and we welcome everyone who is interested in learning more about the cloud, and starting their business, to attend.
In December 2012 we also launched a Machine Learning Research grant program where we fund novel applications of machine learning at cloud scale. We welcome researchers from all over the world to apply and would recommend people reading this to visit.
Thank you for the Interview!