Machines can learn. Incredibly fast. Faster than you. They are getting smarter and smarter every day. They are already changing your world, your business and your life. Artificial intelligence revolution is here. Come and learn how to turn this threat into your biggest opportunity.
This is not another academic conference. Our goal is to foster discussion between machine learning practitioners and all people who are interested in applications of modern trends in artificial intelligence. You can look forward to inspiring people, algorithms, data, applications, workshops and a lot of fun during three days as well as at two great parties.
Reinforcement learning (RL) is a technique that goes way back in the past to early 80s. It became very popular recently because of the use of deep neural networks as the function approximators, in what is called DQN (deep Q-network). The broad application of DQNs ranges from playing ATARI games from raw pixels, playing board game of Go to optimizing large cooling systems in data centers.
In this workshop in the first part we will cover the basic theory of Q-learning and DQN, recent advances and what future development we might expect. In the second part we will go through source codes of RL algorithm and train it on a problem from popular OpenAI gym. The convergence on harder problems can take several days so we will do the training on simpler task that can be done during the workshop.
For this workshop, intermediate level of Python is required.Michal Sustr, Czech Technical University & Milan Le, Charles University
The goal of this workshop is to give attendees a blueprint for building an end to end enterprise ready ML solution and demonstrate its usage on typical ML corporate use cases (telco, digital marketing).
In the first part we will create the base for analyzing the data using Cloudera’s CDH platform. The platform will be setup for data science workloads including integration of necessary tooling. We will also point out best practices and pitfalls to avoid, when running the platform with ML workloads in enterprise environments.
The second session will be dedicated to real-time ingestion and analysis of live data against prepared models on top of our previously created platform.
In the third session will go into detail on some specific machine learning techniques applied in the telco industry use cases like churn prediction or community detection.Johnson Darkwah & Jiri Polcar & Karel Vaculik, Gauss Algorithmic
In malware detection, an extremely low false positive rate is one of the most important targets. This paradigm results in a funnel of multiple different algorithms where each component is designed for achieving a very high precision rate with a potentially low recall. Another specific aspect is that in malware detection, the big data is really big (hundreds of terabytes of files) while posing a challenge of dealing with a very imbalanced class representation (typically much more benign software than malware).
In the first two sessions, we will offer a theoretical look at various algorithms used for malware detections. In the two latter sessions, we will provide the participants with a direct hands-on experience on a real world dataset (more than 200.000 samples). We will also hold a hackathon-like competition in designing a malware detection pipeline that will result in the best recall with the highest precision. For this part of the workshop, a basic level of Python knowledge is recommended (although not obligatory).Martin Balek & Adam Hanka & Monika Seidlova & Ondrej Svec & Tomas Trnka, Avast
Second part of the workshop.Michal Sustr, Czech Technical University & Milan Le, Charles University
Second part of the workshop.Johnson Darkwah & Jiri Polcar & Karel Vaculik, Gauss Algorithmic
Second part of the workshop.Martin Balek & Adam Hanka & Monika Seidlova & Ondrej Svec & Tomas Trnka, Avast
At GoodAI, we believe that that gradual learning is key aspect of AGI. In this hands-on workshop, we shall introduce and motivate the use of gradual learning throughout the lifetime of an agent. We will present a problem which is challenging for the current state of the art deep learning methods, and will demonstrate how it can be solved with the use of gradual learning. We will share code samples for a gradual learner, and work with the audience to construct their own gradual learning agent applicable to both the presented problem, and others.Joseph Davidson & Josef Strunc & Sungmin Aum, GoodAI
Third part of the workshop.Johnson Darkwah & Jiri Polcar & Karel Vaculik, Gauss Algorithmic
Third part of the workshop.Martin Balek & Adam Hanka & Monika Seidlova & Ondrej Svec & Tomas Trnka, Avast
Recent advances in artificial intelligence have been extraordinary. Despite this fact, the measure of our true progress towards human-level artificial intelligence has been elusive. The AI Roadmap Institute has been established to tackle this challenge and push our understanding of the AI landscape beyond our current comprehension. This workshop will offer a glimpse into one of the fundamental tasks of the institute, namely the comparison of disparate roadmaps for building intelligent machines. Such comparison is vital for evaluating where we truly stand relative to our goal, the discovery of potential dead-ends, and the identification of areas of research that the community should really focus on.Jan Feyereisl, GoodAI
In this talk we will overview the topic of image retrieval. Initially, we will focus on retrieval methods based on Bag-of-Words (BoW) image representation that exploit geometric constrains. We discuss how the standard similarity based image retrieval formulation uses BoW with geometry to only satisfy a single user requirement. Then, we show how different formulations employ the same methods to fulfill different requirements in other tasks such as online browsing of image collections, image analysis based on large collections of photographs, or 3D model construction. Finally, we present how BoW with geometry dispenses with the need for costly manual image annotation to perform Convolutional Neural Network (CNN) training.Filip Radenovic, Czech Technical University
Fourth part of the workshop.Martin Balek & Adam Hanka & Monika Seidlova & Ondrej Svec & Tomas Trnka, Avast
The Data Science group at The New York Times develops and deploys machine learning solutions to newsroom and business problems. Re-framing real-world questions as machine learning tasks requires not only adapting and extending models and algorithms to new or special cases but also sufficient breadth to know the right method for the right challenge. I'll first outline how unsupervised, supervised, and reinforcement learning methods are increasingly used in human applications for description, prediction, and prescription, respectively. I'll then focus on the 'prescriptive' cases, showing how methods from the reinforcement learning and causal inference literatures can be of direct impact in engineering, business, and decision-making more generally.Chris Wiggins, New York Times (USA)
We will explore the power of news sentiment to predict financial returns, in particular the returns of a set of European stocks. Building on past decision support work going back to the Delphi method we will describe a text analysis expert weighting algorithm that aggregates the responses of both humans and algorithms by dynamically selecting the best response according to previous performance.
The proposed system is tested through an experiment in which ensembles of experts, crowds and machines analyzed Thomson Reuters news stories and predicted the returns of the relevant stocks mentioned right after the stories appeared. The expert weighting algorithm was better than or as good as the best algorithm or human in most cases.Germán Creamer, Stevens Institute of Technology and Columbia University (USA)
Feed ranking's goal is to provide people with over a billion personalized experiences. We strive to provide the most compelling content to each person, personalized to them so that they are most likely to see the content that is most interesting to them. Similar to a newspaper, putting the right stories above the fold has always been critical to engaging customers and interesting them in the rest of the paper. In feed ranking, we face a similar challenge, but on a grander scale. Each time a person visits, we need to find the best piece of content out of all the available stories and put it at the top of feed where people are most likely to see it. To accomplish this, we do large-scale machine learning to model each person, figure out which friends, pages and topics they care about and pick the stories each particular person is interested in. In addition to the large-scale machine learning problems we work on, another primary area of research is understanding the value we are creating for people and making sure that our objective function is in alignment with what people want.Lars Backstrom, Facebook (USA)
TensorFlow is an open-source Python library for machine learning and neural networks. With this new tool, deep machine learning transitions from an area of research into the realm mainstream software engineering. In this session, learn how you can leverage TensorFlow to make maintainable, high performance machine learning systems.Yufeng Guo, Google (USA)
We will describe methods for designing feedforward and recurrent neural architectures and their effective applications to diverse problems in the life sciences. In particular, we will demonstrate deep learning applications for the prediction of the properties of small and large molecules (proteins) in bioinformatics, the analysis of circadian rhythms in systems biology, and the detection of cancer and other pathologies in biomedical imaging.Pierre Baldi, University of California (USA)
A key reason for giving a machine intelligence is to enable it to handle new unforeseen challenges. A system that can do this autonomously over extended periods of time meets an important requirement for what we call general intelligence. Predictions and promises notwithstanding, machines with general intelligence and high levels of autonomy have been slow in coming, and pprinciples for how to design such systems have been missing. In this talk I will present new methodological and engineering principles for designing a machine that becomes increasingly better at behaving in underspecified circumstances with experience, in a goal-directed way, continuously, on the job, with minimal up-front information. Starting from a tiny amount of designer-specified code – a seed – the machine recursively self-improves, ultimately growing its knowledge to large multiples of the seed. I will show an example of how such a machine can learn a complex task such as a TV-style interview. Knowing nothing about sentence structure or grammar, interviews, multimoal gestures, or turntaking, it learns in an unsupervised manner how to effortlessly couple and coordinate multimodal gesture and speech, how to construct complex grammatically correct sentences, to correctly interpret complex spoken utterances, resolve anaphora, interpret multimodal gestures, take turns smoothly to achieve the goal of cooperative dialogue, and properly and appropriately structure a whole interview in live interaction with a human.Kristinn R. Thórisson, Reykjavik University & Icelandic Institute for Intelligent Machines, (IS)
Many scientists, futurologists and philosophers have predicted that humanity will achieve a technological breakthrough and create Artificial General Intelligence (AGI). It has been suggested that AGI may be a positive or negative factor in the global catastrophic risk. In order to mitigate a dangerous AGI system it is important to understand how the system came to be in such a state. In this talk, I will survey, classify and analyze a number of pathways, which might lead to arrival of dangerous AGI.Roman Yampolskiy, University of Louisville (USA)
Wojciech will provide some tips on right tools, scope of analytical process, reusable components and creating functions that leads to development-friendly environment for creating specific machine learning pipelines.Wojciech Indyk, Konica Minolta Research Lab Europe (PL)
Multi-instance learning applies to problems where samples are described by a set of vectors of arbitrary size. Such problems naturally occur in text analysis, image recognition, analysis of molecules, and network security. In this talk, we will first show how neural networks can be elegantly applied to MIL problems. Then we will move to modelling more complex tree structures describing structure of network traffic.Tomas Pevny, Cisco (CZ)
Our database contains hundreds of terabytes of executable files which can be used for training machine learning models. We introduce a flexible, in-house developed machine learning system for partitioning the space of fingerprints of executable files that automatically trains separate local models in each partial volume leveraging both CPU and GPU training of random forests.
The system allows for a fast prototyping of new features both on local models (on order of seconds) and in case of success, also on the entire dataset (on order of hours), thereby giving our analysts a powerful tool to engineer new ways to stop malware.Martin Vejmelka, Avast (CZ)
First 50 tickets
Before April 17, 2017
From April 18, 2017
The registration fees cover the official program, coffee breaks, lunches and parties.
Most of the talks will be given by well-known invited speakers. However, we would also like to give students and startups an opportunity to present their work and research. It will be realized in the form of short lightning talks. Each speaker will get a free ticket to the conference.
If you are interested in becoming a lightning speaker, please send us a short paper (less than 1000 words) describing your machine learning project. Our committee will select up to 10 speakers for the lightning session.
Submission deadline: February 28, 2017
Notification: March 15, 2017
We understand that students have usually limited budget for conference fees. That is why we offer several volunteer positions. We are looking for conference volunteers, an experienced photographer and a camera operator. If you are interested in helping us and getting a free ticket, please contact us.
If you have any questions about Machine Learning Prague, please
Copyright © 2017 Machine Learning Prague s.r.o.
Blanicka 589/5, Vinohrady, 120 00 Praha 2