2018 Machine Learning Predictions from the Experts Themselves
Our vast experience with planning data science conferences across a multitude of industries has enabled us to host, listen and learn valuable insights into the industry’s most ambitious goals and research advancements. As the data science community heads towards 2018, we asked our top speakers to comment on 2017’s most impactful achievements in Artificial Intelligence and make a few predictions for 2018. We summarize the most notable insights in this post, and offer expert commentary on the advancements, predictions and lessons learned regarding machine learning algorithms and deep learning systems.
Daniel Monistere, SVP-Client Solutions at Nielsen points out the technology advancement electronic devices have met and the increase in their storage and data gathering capabilities. Also, applications have become intelligent being able to collect user data. However, he stresses the importance of correctly representing the usage data, measuring and describing accurately the phenomenon of interest.
Another upwarding trend is the commoditization of mobile deep learning availability. Keisuke Inoue, VP of Data Science at Emogi, mentions that technology giants such as Facebook, Google and Apple have already released mobile versions of deep learning platforms in November 2016, May 2017 and June 2017 respectively. These are Cafe2Go (Facebook), Tensorflow Lite from Google and CoreML from Apple which allows developers to integrate a wide variety of machine learning models to iOS and Mac OS platforms.
Computing Power and Storage Capacity
Kevin Perko, Data Science Lead at Scribd acknowledges the merits of rising GPU compute. GPU enables faster training of artificial neural networks in a matter of hours, instead of weeks. In the following years, he argues that technological improvements of such magnitude over existing systems will take over the query and visualization layers which will further extend in the analytic flow. Sharing from his professional experience, Perko explains that he uses deep learning algorithms to improve the search experience of users by deepening the system’s understanding of intent. This will further enable Scribd to update the recommendations system in such ways to take advantage of its innate ability to automate feature engineering and manage complex relationships in the data.
Computing power and storage capacity continue to grow exponentially while their costs are decreasing. Therefore, they do not remain confined within the world’s largest corporations sphere, but become increasingly accessible to startups as well. Therefore, any company can become a data company with affordable means of gathering, storing and computing huge volumes of data. 
Industry Lengths and APIs
Joe Devon, Founding Partner at Diamond points out that despite the lack of ML experts on the job market, developers are able to send their data and receive competent results using already built APIs from the major cloud providers. A good example is Google’s Professional Services team whose main purpose is to train companies so they can use Google’s API. He also discusses the rapid spread of Machine Learning systems into unexpected fields such as Legal Contracts, Medicine or Film industry. Examples of industries that adopted machine learning to improve the work they deliver do not stop here.
Education is seeking to improve schooling experience by implementing machine learning programs which allow for faster identification of learning disabilities. Improved curriculum, smartly generated assignments which meet the right combination of vocabulary and questions to demonstrate comprehension on any subject or spotting misconceptions and gaps in students’ knowledge are only a few illustrations of how machine learning and AI can contribute.
How about agriculture? Descartes Labs use machine learning to analyze an immense amount of satellite imagery data in order to produce vastly more accurate crop yield predictions. This will be of great help when global food shortages are addressed.
Another interesting application of machine learning algorithms is conservation. The Nature Conservancy organization has as a main goal to ‘conserve the lands and waters on which all life depends’.  These systems will learn to do everything from mapping birds populations, monitoring the health of tuna fisheries or fighting against global warming.
Wanderers in a Black Box
Michael Housman, Co-Founder and Chief Data Science at RapportBoostAI is concerned with the fact that data scientists face a loss of intuition when working with highly predictive machine learning models. Even though the statistical results are fantastic, one should be able to question the how and why behind them. Once these self-learning machines are trained they show concerning behaviour by adding an additional layer of complexity and opaqueness in the process of generating results. Further, this makes it very difficult to discover and fix particular traits and raises questions regarding the autonomy, decision-making and responsibility of AI algorithms. The black box problem which comes with the amazing advancement of the ML approaches should be taken into consideration if the results are used to drive action.
Jonathan Morra, VP and Data Science at ZEFR stresses the increasing availability of user-friendly data science tools which in the future will act as a divide between those who know how to use them and those who understand the math behind them. This will further reflect down on the methods used when tackling various problems as some are better solved through algorithms versus neural networks. He predicts a small but noticeable backlash to deep learning becoming synonymous with machine learning.
K. Inoue also points out that it becomes easier for programmers to use pre-trained machine learning models such as Inception-v3 (Google) and commercial APIs without having to run their own machine learning experiments. This might undermine the scientists’ motivation to methodologically contribute while some areas still remain unexplored. He gives as example the unavailability of models or APIs which recognize creative messaging content — animated stickers or GIFs.
However, Becky Tucker, Senior Data Scientist at Netflix states that there will be a shift away from strict focus on machine learning to a broader view of how machine learned models integrate into existing workflows, tools, and applications. Therefore, the qualities needed for successful data science teams are adoption and usability.
Product and Content Optimization
Speaking from her experience in the media and entertainment industry, Hollie Choi, Executive Director of IT Intellectual Property Management at 20th Century Fox, observes the emerging trends around digital assets and metadata management. With the aid of predictive analytics, new products will be developed to meet people’s needs and expectations.
When discussing content optimization one should take into consideration the initial purpose of machine learning algorithms, recommender systems. Over time, these types of algorithms became a staple of the internal processes within a certain company which aid the creators and decision makers in the production of content, asserts Ali Baghshomali, Buzzfeed, Data Scientist.
Alejandro Cantarero, VP of Data at Los Angeles Times Media Group, argues that adding structured elements to text data and learning how those elements connect represents a major area of research in 21st century journalism. Techniques such as event-centric knowledge graphs are believed to impact greatly on many aspects of media industry such as content recommendation, search and discovery, and automated storytelling.
Cantarero explains that conversational interfaces play an important role in newsrooms around the world. Apart from the emerging tools to distribute and consume news such as chatbots or voice processors, these human-computer interactions will help develop an easier understanding of analytics, business insights and machine learning algorithms outputs for any business employee comprehension.
Hollie Choi predicts that new machine learning systems will be created to efficiently deal with and detect abnormal behavior of transactional data. Training machines to process and analyze threat data brings enormous benefits for information security in organizations. Two examples are the scale of data which AI systems can work with and its efficiency in structuring it to spot the threat in short time. Jonathan Morra (ZEFR) foresees an increased focus on productionalization of machine learning pipelines including data Quality assurance, monitoring, reporting and CI/CD for machine learning models at scale.
Attention is being brought to the potential threats AI can pose to this world. Before rushing into thinking about conscious self-healing robots conquering the human race, one should first consider people who seek to harm others. Joe Devon talks about drones or self-driving technology anonymously paid for by blockchains which become alarmingly dangerous in terrorists’ hands.