Odsc machine-learning-guide-v1
0 likes43 views
This document provides a summary of popular machine learning resources from 2018, including the top 10 blogs and talks from the Open Data Science community. It highlights how machine learning grew significantly in 2018, drawing more investment and attention across industries. Practitioners paid increased attention to the societal impacts of their work and focused on technical approaches to address bias, lack of transparency, and other issues. The resources provided are meant to help readers further their understanding of machine learning and shape how these tools are applied.
Odsc machine-learning-guide-v1
- 1. Machine Learning Guide 20 The Open Data Community’s Top Resources
- 2. 2018 was another year of intense growth - and attention - for the fields of machine learning and artificial intelligence. As technologies and techniques improve across industry and academia, this space draws more investment, but also more scrutiny. As a result, in 2018 we saw the community of practitioners paying more attention to the impacts of their work and research on the wider world. All corners of the field saw a focus on technical approaches to resolv- ing bias, moving away from operational black boxes, and taking up the ideas of transparency and explainability. Now is the time to ask and answer these questions. The power of the tools and applications built from the research in the field is clear; we must now consider the world we wish to shape using these tools. They are excellent magnifying glasses, capable of revealing intricate patterns in the information all around us. They have the ability to help us bet- ter understand the world that we have, and more effectively build the world that we want. But like all tools, we must handle them carefully or we risk harming ourselves and others. We risk further reinforcing the walls we find in our society today built during a less informed and analytical era. The only way to combat this is to widen the conversation, to equip anyone who wants to learn with an understanding of the fundamental principles and technical knowledge of the field. Only by this sharing of knowledge, this open exchange of ideas, will we be able to make sure these tools truly benefit everyone. The work that was done in 2018 and that will be done in 2019 will define not just the machine learning com- munity, but how the world uses these technologies in our day-to-day lives. In this report, you will find the 10 most popular talks from the 2018 Open Data Science Conferences and the 10 most popular blogs from OpenDataScience.com in 2018. I hope you take the information and inspiration herein and use it to further your journey into machine learning. Please share it widely with anyone you know who wants to be a part of the community of data scientists and engineers that will help shape the future of data science and machine learning. Katherine Gorman Executive Producer and Co-host of Talking Machines & Executive Producer, Collective Next Contents 4 top blogs 8 from the experts 9editor’s note6top ODSC talks 10what’s next FOREWORD
- 3. Top Blogs | Page 4 TOP OPEN DATA SCIENCE BLOGS Machine Learning Anthology | 2018 10 Client-side Web Development and Machine Learning, Caspar Wylie See how and why machine learning and web development are beginning to collaborate rather successfully, such as through the collaboration with JavaScript. Read it here. In 2018 alone, we published nearly 400 articles on data science. Machine learning was a common and well-received topic in our community. Here are the 10 most-read machine learning blogs. Machine Learning for Beginners — a How-to Guide, Spencer Norris Check out this tutorial to get a general outline of how different popular machine learning algorithms work, plus some code recipes you can use if you want to experiment. Read it here. How to Define a Machine Learning Problem Like a Detective, Spencer Norris What steps do you need to take to identify the best machine learning problems to ask? Learn how to define a machine learning problem, but through the eyes of a detective. Read it here. Efficient, Simplistic Training Pipelines for GANs in the Cloud with Paperspace, Spencer Norris Generative adversarial networks are making waves in the world of machine learning. Learn about a package that wraps PyTorch implementations for ten different types of GANs in an easy-to-use interface. Read it here. Crash Course: Pool-Based Sampling in Active Learning, Spencer Norris Active learning is a class of machine learning problems where labeled data isn’t available for supervised algorithms. Read more about one of the most common active learning problems in the field: pool-based sampling. Read it here. Gain insights into the three primary types of gradient descent, the most commonly used optimization method deployed in machine learning and deep learning algorithms. Read it here. Understanding the Three Primary Types of Gradient Descent, Daniel Gutierrez This recap of the reinforcement learning workshop from ODSC London 2018 will give you the key takeaways you need to implement the framework yourself. Read it here. An Introduction to Reinforcement Learning Concepts, Diego Arenas Learn the basics behind gradient boosting for statistical learning and the popular XG- Boost implementation — the most recent evolution of gradient boosting. See more about how XGBoost became king of the hill for data scientists desiring accurate predictions. Read it here. Gradient Boosting and XGBoost, Daniel Gutierrez Overfitting is an interesting problem with fascinating solutions embedded in the very structure of the algorithms you’re using. Here, we break down what overfitting is and how we can provide an antidote to it in the real world. Read it here. What Overfitting is and How to Fix It, Spencer Norris Unsupervised machine learning can be very powerful in its own right, and clustering is by far the most common expression of this group of problems. Check out this quick rundown of three of the most popular clustering approaches and what situations each is best suited to. Read it here. Three Popular Clustering Methods and When to Use Each, Spencer Norris
- 4. Top Talks | Page 6 Machine Learning Anthology | 2018 Out of 300+ talks from ODSC conferences in 2018, here are the 10 top-rated sessions covering machine learning. TOP TALKS FROM ODSC CONFERENCES 10 Andreas Mueller Introduction to Machine Learning & Intermediate Machine Learning with scikit-learn Start with the basics of machine learning before you dive into the tools, applications, and advanced functions! Watch part one here. Watch part two here. Bernard Marr Artificial Intelligence and Machine Learning in Practice This ODSC Europe keynote features Bernard Marr, internationally best-selling business author and strategic advisor to companies and governments. Gain insight into applied AI and machine learning for your organization. Watch here. Kirk Borne A Tour of Machine Learning Algorithms: The Usual Suspects in Some Unusual Applications Walk through use cases for several machine learning algorithms to see how to implement them in interesting and unexpected ways. Topics include predictive modeling and anomaly discovery. Watch here. Jon Peck OS for AI: How Serverless Computing Enables the Next Gen of ML This talk examines the need for and implementations of an “Operating System for AI” — a common interface to use and combine algorithms and a general ar- chitecture for serverless machine learning that is discoverable, versioned, scal- able, and shareable. Watch here. Jared Lander Machine Learning in R Parts I & II This two-part course focuses on the available methods for implementing machine learning algorithms in R and examines some of the underlying theory behind the curtain. Watch part one here. Watch part two here. Randy Olson The Past, Present, and Future of Automated ML In this talk, Randy Olson draws from his AutoML research to discuss the benefits of AutoML and highlights some promising future directions of the field. Watch here. Jeffrey Yau Multivariate Time Series Forecasting Using Statistical and ML Models This lecture discusses the formulation Vector Autoregressive (VAR) Models, one of the most important classes of multivariate time series statistical models, and neural network-based techniques. Watch here. Yuriy Guts Target Leakage in Machine Learning Target leakage is one of the most difficult problems in developing real-world machine learning models. Hear more about real-life examples of data leakage at different stages of the data science project lifecycle, and discuss various counter- measures and best practices for model validation. Watch here. Crissman Loomis Machine Learning in Chainer Python When choosing a framework for working on neural networks, it is important to choose a framework that is flexible and allows for customization. Chainer is a neural network framework written almost entirely in Python. Gain the knowledge you need to get started with Chainer, from data formatting and augmentation to reinforcement learning and more. Watch here. Keith Santarelli & Eric Schles Making the Most of Your Time Series: Signal Processing for Machine Learning applications Learn more about common machine learning signal processing techniques on time series data. This talk discusses a number of common tools in signal processing and shows how they can be implemented in various Python packages, including tools to remove “noise” to find underlying trends. Watch here.
- 5. From the Experts | Page 8 Machine Learning Anthology | 2018 FROM THE EXPERTS: My hope is that new practitioners will start to have a clearer idea of how to break into the field. Currently there is no standardization across different types of data science roles and titles. In addition, since the field is multidisciplinary, those trying to learn have a difficult time prioritizing what they should learn and don’t feel confident in starting to apply to jobs because the learning could theoretically go on forever. My message to those people is to start applying so that they can receive feedback from the job market to help them in their pursuit. Kristen Kehrer Founder, Data Moves Me LLC Almost all ML models are based on statistics. No one, including your customers and employees, likes to be treated as a statistic, which is what your ML processes will tend to do. Plan to mitigate that from the beginning. Adam Breindel Independent Consultant and Instructor, ML/AI and Data Engineering 2018 was the year that machine learning, and by extension deep learning, have taken many industries by storm. Machine learning has touched just about every problem domain, bringing accelerated diversity in the types of problems being solved. Daniel Gutierrez Data Science Consultant One common thread I’ve seen at every stop in my career is the ever-increasing focus on using data to make informed decisions. Now, perched in a place to see day-to-day developments in the field, I’ve learned so much about various frameworks and languages, the unique ways that different industries are using AI, and of course, how strong the open data science community is in sharing its knowledge. There are definitely a few topics in particular that I’m paying extra attention to heading into 2019. Apache Spark’s MLlib is a frequent topic of conversation when I speak with data science experts, largely thanks to MLib’s remarkable scaling ability in implementing multiple ML algorithms. At this rate, I can see the scikit-learn library for Python becoming a common job requirement for ML experts — it’s appearing in social feeds, featured blogs, and job listings that cross my desk more and more. The TensorFlow framework is popping up in countless ML tutorials, so I’m hoping people stay active with it, even with newer libraries releasing frequently. This should be a no-brainer, but start looking into automated machine learning if you haven’t already, as it will help to increase the pace in which you can create more complex ML processes and algorithms. Let the machines work for you! Data scientists need to make 2019 a strong year to address the common “black box” problem. As Daniel Gutierrez, a data science consultant and frequent author for OpenDataScience.com put it, “I hope the trend of ‘explainability’ or ‘interpretability’ of AI will continue to be seen as critical to the continued acceptance of the technology.” Since machine learning may open up more chances for vulnerabilities to appear, developers need to be careful with their applications and to avoid leaving entryways for malicious attacks, and to be well- prepared to defend against potential adversarial attacks. It’s better to spend the time building up your defenses rather than going through the headache of resolving issues from hackers and malware. Whether you use the information provided in this anthology to learn a new tool, framework, or you’re now more invested in security and defense, I hope that all of the videos, blogs, and insights from the experts in this anthology prove useful for you. Whether you’re new to data science and machine learning or a seasoned vet, already working in applied data science or academia, or you’re just a fan of new technology, the open data science community is a good place to share knowledge and expand understanding of the most exciting topics in applied data science Alex Landa, ODSC content manager Letter from the Editor “ To be successful, machine learning adopters must enable a flexible infrastructure and be agile. It is critical to experiment and accept failure in exchange for quicker learning and less money spent going after projects that aren’t successful. Kate Strachnyi Data Visualization Specialist; Host of Humans of Data Science; Author of “The Disruptors: Data Science Leaders and Journey to Data Scientist” Avoid learning everything at once. Content from the ML/data science community can make you feel like you’ve got to know everything, like yesterday. Resist this. Don’t pick up a textbook and read it cover to cover. If you’re interested in getting started, spend time thinking about one question from one subject area that you’re interested in. Find some data, clean it, explore it, and use it to answer your questions. You won’t know how. Research what you need to know and figure it out, step by step. Brandon Dey Technical Team Lead, Data Science Global Marketing, Fisher Investments What do leading data scientists think about the state of machine learning? Open Data Science HIGHLIGHTS Conferences in 5countries blog on @ODSC Medium: Three Popular Clustering Methods and When to Use Each, Spencer Norris #1 70,000subscriptions to our weekly newsletter Learn.AI course with the HIGHEST ENROLLMENT: Machine Learning and NLP for Detecting Fake News Watch here
- 6. There are many ways you can engage with the Open Data Science Community today! BE A PART OF THE ODSC COMMUNITY ODSC Events East 2019: April 30-May 3 India 2019: August 7-10 West 2019: October 29-November 1 Europe 2019: November 19-22 Meetups We hold meetups in 37 cities around the world, designed to convene data scientists for education, networking, and even a little fun. See upcoming events here. Weekly Newsletter Don’t miss any future articles on data science and machine learning! Sign up for our weekly newsletter and get tutorials, insights, and the latest news sent to you directly. Webinars We offer free webinars several times a month, covering a variety of topics. Follow this page to learn more about upcoming webinars. Become a Speaker Are you a technical or business expert in the world of data science and AI? Consider speaking at one of our events! Each event has its own speaker submission page: ODSC East 2019 ODSC Europe 2019 And more coming soon! Partner with ODSC We also offer opportunities for partnerships! Have your product, service, or research seen by thousands of data scientists at an event. Learn more here.