Automated Techniques Could Make It Easier to Develop AI
Machine-learning researchers make many decisions when designing new models. They decide how many layers to include in neural networks and what weights to give inputs to each node. The result of all this human decision-making is that complex models end up being “designed by intuition” rather than systematically, says Frank Hutter, head of the machine-learning lab at the University of Freiburg in Germany.
A growing field called automated machine learning, or autoML, aims to eliminate the guesswork. The idea is to have algorithms take over the decisions that researchers currently have to make when designing models. Ultimately, these techniques could make machine learning more accessible.
Although automated machine learning has been around for almost a decade, researchers are still working to refine it. Last week, a new conference in Baltimore—which organizers described as the first international conference on the subject—showcased efforts to improve autoML’s accuracy and streamline its performance.
There’s been a swell of interest in autoML’s potential to simplify machine learning. Companies like Amazon and Google already offer low-code machine-learning tools that take advantage of autoML techniques. If these techniques become more efficient, they could accelerate research and allow more people to use machine learning.
The idea is to get to a point where people can choose a question they want to ask, point an autoML tool at it, and receive the result they are looking for.
That vision is the “holy grail of computer science,” said Lars Kotthoff, a conference organizer and assistant professor of computer science at the University of Wyoming. “You specify the problem, and the computer figures out how to solve it—and that’s all you do.”
But first, researchers will have to figure out how to make these techniques more time and energy efficient.
What is autoML?
At first glance, the concept of autoML might seem redundant—after all, machine learning is already about automating the process of gaining insights from data. But because autoML algorithms operate at a level of abstraction above the underlying machine-learning models, relying only on the outputs of those models as guides, they can save time and computation. Researchers can apply autoML techniques to pre-trained models to gain fresh insights without wasting computation power repeating existing research.
For example, research scientist Mehdi Bahrami and his coauthors at Fujitsu Research of America presented recent work on how to use a BERT-sort algorithm with different pre-trained models to adapt them for new purposes. BERT-sort is an algorithm that can figure out what is called “semantic order” when trained on data sets—given data on movie reviews, for example, it knows that “great” movies rank higher than “good” and “bad” movies.
With autoML techniques, the learned semantic order can also be extrapolated to classifying things like cancer diagnoses or even text in the Korean language, cutting down on time and computation.
“BERT takes months of computation and is very expensive—like, a million dollars to generate that model and repeat those processes,” Bahrami said. “So if everyone wants to do the same thing, then it’s expensive—it’s not energy efficient, not good for the world.”
Although the field shows promise, researchers are still searching for ways to make autoML techniques more computationally efficient. For example, methods like neural architecture search currently build and test many different models to find the best fit, and the energy it takes to complete all those iterations can be significant.
AutoML techniques can also be applied to machine-learning algorithms that don’t involve neural networks, like creating random decision forests or support-vector machines to classify data. Research in those areas is further along, with many coding libraries already available for people who want to incorporate autoML techniques into their projects.
The next step is to use autoML to quantify uncertainty and address questions of trustworthiness and fairness in the algorithms, says Hutter, a conference organizer. In that vision, standards around trustworthiness and fairness would be akin to any other machine-learning constraints, like accuracy. And autoML could capture and automatically correct biases found in those algorithms before they’re released.
The search continues
But for something like deep learning, autoML still has a long way to go. Data used to train deep-learning models, like images, documents, and recorded speech, is usually dense and complicated. It takes immense computational power to handle. The cost and time for training these models can be prohibitive for anyone other than researchers working at deep-pocketed private companies.
One of the competitions at the conference asked participants to develop energy-efficient alternative algorithms for neural architecture search. It’s a considerable challenge because this technique has infamous computational demands. It automatically cycles through countless deep-learning models to help researchers pick the right one for their application, but the process can take months and cost over a million dollars.
The goal of these alternative algorithms, called zero-cost neural architecture search proxies, is to make neural architecture search more accessible and environmentally friendly by significantly cutting down on its appetite for computation. The result takes only a few seconds to run, instead of months. These techniques are still in the early stages of development and are often unreliable, but machine-learning researchers predict that they have the potential to make the model selection process much more efficient.
ISASecure Program Announces New ISASecure Certification Offering
Sept. 1, 2022 – The ISASecure program is announcing the new ISASecure certification offering for the industrial internet of things (IIoT) components based on the ISA/IEC 62443 series of standards.
The IIoT Component Security Assurance (ICSA) certification was inspired by recommendations published in the joint ISA Global Security Alliance (ISAGCA) and ISA Security Compliance Institute (ISCI) study. Details of this landmark study are available in the Learning Center section of the ISASecure website and were presented during our October 2021 webinar. The study and resulting ISASecure IIOT certification scheme address the urgent need for industry-vetted IIoT certification programs.
Join us on Sept. 7, 2022, at 11 a.m. ET for a live webinar where we will be presenting this important new certification offering. This webinar will provide an overview of the new ISASecure IIOT Device and Gateway certification program and its basis in the ISA/IEC 62443 set of industry standards. Register here.
About ISASecure
Founded in 2007, the ISA Security Compliance Institute’s (ISCI) mission is to provide the highest level of assurance possible for the cybersecurity of automation control systems. ISCI has been conducting ISASecure certifications on automation and control systems since 2011 through its network of ISO/IEC 17065 accredited certification bodies.
The Institute was established by thought leaders from major organizations in the automation controls community seeking to improve the cybersecurity posture of critical infrastructure for generations to come. Prominent ISASecure supporters include Chevron, ExxonMobil, Saudi Aramco, Shell, Honeywell, Schneider Electric, TUV Rheinland, Yokogawa, YPF, exida, GE Digital, Synopsis, CSSC, CSA Group, IPA-Japan, and others.
The Institute’s goals are realized through ISASecure compliance programs, education, technical support, and improvements in suppliers’ development processes and users’ life cycle management practices. The ISASecure designation ensures that automation products conform to industry consensus cybersecurity standards such as ISA/IEC 62443, providing confidence to users of ISASecure products and systems and creating product differentiation for suppliers conforming to the ISASecure specification.
About the International Society of Automation (ISA)
The International Society of Automation (ISA) is a non-profit professional association founded in 1945 to create a better world through automation. ISA advances technical competence by connecting the automation community to achieve operational excellence and is the trusted provider of standards-based foundational technical resources, driving the advancement of individual careers and the overall profession. ISA develops widely used global standards; certifies professionals; provides education and training; publishes books and technical articles; hosts conferences and exhibits; and provides networking and career development programs for its members and customers around the world.
Robot Sales Hit Record High in North America for Third-Straight Quarter
Aug. 29, 2022 – For the third-straight quarter, robot sales in North America hit a record high, driven by a resurgence in sales to automotive companies and an ongoing need to manage increasing demand to automate logistics for e-commerce. According to the Association for Advancing Automation, of the 12,305 robots sold in Q2 2022, 59% of the orders came from the automotive industry with the remaining orders from non-automotive companies largely in the food & consumer goods industry, which saw a 13% increase in unit orders over the same period, April through June, in 2021.
According to A3, 59%25 of the orders in Q2 2022 came from the automotive industry with the remaining orders from non-automotive companies largely in the food & consumer goods industry.
“While automotive entities have long been the frontrunner in deploying robotics and automation, the last few years have seen food & consumer goods, life sciences, and other industries grow at even higher rates,” said A3 President Jeff Burnstein. “While this quarter shows a marked shift back to historic norms with more robots going to automotive than to any other industry, the continued growth of robotics in food & consumer goods companies especially demonstrates the ongoing need to automate warehouse logistics for handling the exploding growth of e-commerce. We’re excited to share the latest on robots in the logistics space at our upcoming Autonomous Mobile Robots & Logistics Week in Boston in October.”
The 12,305 units sold in Q2 2022 is 25% more than sold in the same period in 2021 and 6% more than sold in the first quarter of 2022, which saw 11,595 robots sold. The Q2 2022 value of $585 million is the second-best quarter ever for revenue, down 9% from the previous record quarter—Q1 2022, which saw $646 million in revenue. When combined with 2022’s first quarter results, the previous record, the North American robotics market is off to its best start ever, with 23,903 robots ordered at a value of $1.249 billion. The market grew 26% and 29% for units ordered and revenue, respectively, over 2021.
A record fourth quarter in 2021 resulted in the strongest year ever for North American robot sales, with 39,708 units sold at a value of $2 billion, and 2022 is on pace for another record year. Alex Shikany, Vice President – Membership & Business Intelligence, A3, will discuss the end-of-year numbers in detail at the next A3 Business Forum in January.
“The larger trend towards robots being used to benefit more companies in North America continues,” Burnstein added. “This makes it critical to educate system integrators and users now about how to deploy robots while keeping workers safe. Our International Safety Robotics Conference (ISRC) will specifically address the most up-to-date safety standards, providing the best practices and use cases that will help all companies safely succeed with automation.”
Register for A3’s Educational Conference now
In addition to ISRC, scheduled for Sept. 27-29 in Columbus, Ohio, A3 will hold the Artificial Intelligence & Smart Automation Conference, a one-day event to help those interested start their journey to unlock the power of AI, with discussions on data strategy, advances in AI robotics and machine vision, and AI-powered optimization and prediction. The conference will take place Sept. 29, also in Columbus.
AMR & Logistics Week, scheduled for Oct. 10-13 in Boston, will be co-located with The Vision Show, designed to provide the right solution providers, the right technology, and the right expertise to implement vision and imaging systems.
A3’s Business Forum, January 16-18, 2023, in Orlando, Florida, an annual networking event for robotics, vision & imaging, motion control & motors, and artificial intelligence industry professionals, will be followed by The Automate Show (May 22-25, 2023, in Detroit), the largest and most inspiring showcase of automation in North America.
About Association for Advancing Automation (A3)
The Association for Advancing Automation (A3) is the leading global advocate for the benefits of automating. A3 promotes automation technologies and ideas that transform the way business is done. Members of A3 represent nearly 1,100 automation manufacturers, component suppliers, system integrators, end users, academic institutions, research groups, and consulting firms that drive automation forward worldwide.
A3 hosts a number of industry-leading events, including the International Robotics Safety Conference (Sept. 27-29, in Columbus), the AI & Smart Automation Conference (Sept. 29, also in Columbus), Autonomous Mobile Robots & Logistics Week (Oct. 10-13, in Boston), The Vision Show (Oct. 11-13, also in Boston), A3 Business Forum (Jan. 16-18, 2023, in Orlando) and the Automate Show (May 22-25, 2023, in Detroit)
Artificial intelligence has come a long way since scientists first wondered if machines could think.
In the 20th century, the world became familiar with artificial intelligence (AI) as sci-fi robots that could think and act like humans. By the 1950s, British scientist and philosopher Alan Turing posed the question “Can machines think?” in his seminal work on computing machinery and intelligence, where he discussed creating machines that can think and make decisions the same way humans do (reference 1). Although Turing’s ideas set the stage for future AI research, his ideas were ridiculed at the time. It took several decades and an immense amount of work from mathematicians and scientists to develop the field of artificial intelligence, which is formally defined as “the understanding that machines can interpret, mine, and learn from external data in a way that imitates human cognitive practices” (reference 2).
Even though scientists were becoming more accustomed to the idea of AI, data accessibility and expensive computing power hindered its growth. Only when these challenges were mitigated after several “AI winters” (with limited advances in the field) did the AI field experience exponential growth. There are now more than a dozen types of AI being advanced (figure).
Areas within AI (reference 3)
Due to the accelerated popularity of AI in the 2010s, venture capital funding flooded into a large number of startups focused on machine learning (ML). This technology centers on continuously learning algorithms that make decisions or identify patterns. For example, the YouTube algorithm may recommend less relevant videos at first, but over time it learns to recommend better-targeted videos based on the user’s previously watched videos.
The three main types of ML are supervised, unsupervised, and reinforcement learning. Supervised learning refers to an algorithm finding the relationship between a set of input variables and known labeled output variable(s), so it can make predictions about new input data. Unsupervised learning refers to the task of intelligently identifying patterns and categories from unlabeled data and organizing it in a way that makes it easier to discover insights. Lastly, reinforcement learning refers to intelligent agents that take actions in a defined environment based on a certain set of reward functions.
Deep learning, a subset of ML, had numerous ground-breaking advances throughout the 2010s. Similar to the connections between the nervous system cells in the brain, neural networks consist of several thousand to a million hidden nodes and connections. Each node acts as a mathematical function, which, when combined, can solve extremely complex problems like image classification, translation, and text generation.
Impact of artificial intelligence
Human lifestyle and productivity have drastically improved with the advances in artificial intelligence. Health care, for example, has seen immense AI adoption with robotic surgeries, vaccine development, genome sequencing, etc. (reference 5). So far, the adoption in manufacturing and agriculture has been slow, but these industries have immense untapped AI possibilities (reference 6). According to a recent article published by Deloitte, the manufacturing industry has high hopes for AI because the annual data generated in this industry is thought to be around 1,800 petabytes (reference 7).
This proliferation in data, if properly managed, essentially acts as a “fuel” that drives advanced analytical solutions that can be used for the following (reference 8):
becoming more agile and disruptive by learning trends about customers and the industry ahead of competitors
saving costs through process automation
improving efficiency by identifying processes’ bottlenecks
enhancing customer experience by analyzing human behavior
making informed business decisions, such as targeted advertising and communication (reference 9).
Ultimately, AI and advanced analytics can augment humans to help mitigate repetitive and sometimes even dangerous tasks while increasing focus on endeavors that drive high value. AI is not a far-fetched concept; it is already here, and it is having a substantial impact in a wide range of industries. Finance, national security, health care, criminal justice, transportation, and smart cities are examples of this.
AI adoption has been steadily increasing. Companies are reporting 56 percent adoption in 2021, an uptick of 6 percent compared to 2020 (reference 10). With technology becoming more mainstream, the trends of achieving solutions that emphasize “explainability,” accessibility, data quality, and privacy are amplified.
“Explainability” drives trust. To keep up with the continuous demand for more accurate AI models, hard-to-explain (black-box) models are used. Not being able to explain these models makes it difficult to achieve user trust and to pinpoint problems (bias, parameters, etc.), which can result in unreliable models that are difficult to scale. Due to these concerns, the industry is adopting more explainable artificial intelligence (XAI).
According to IBM, XAI is a set of processes and methods that allows human users to comprehend and trust the ML algorithm’s outputs (reference 11). Additionally, explainability can increase accountability and governance.
Increasing AI accessibility. The “productization” of cloud computing for ML has taken the large compute resources and models, once reserved only for big tech companies, and put them in the hands of individual consumers and smaller organizations. This drastic shift in accessibility has fueled further innovation in the field. Now, consumers and enterprises of all sizes can reap the benefits of:
pretrained models (GPT3, YOLO, CoCa [finetuned])
building models that are no-code/low-code solutions (Azure’s ML Studio)
serverless architecture (hosting company manages the server upkeep)
instantly spinning up more memory or compute power when needed
improved elasticity and scalability.
Data mindset shift. Historically, model-centric ML development, i.e., “keeping the data fixed and iterating over the model and its parameters to improve performances” (reference 12), has been the typical approach. Unfortunately, the performance of a model is only as good as the data used to train it. Although there is no scarcity of data, high-performing models require accurate, properly labeled, and representative datasets. This concept has shifted the mindset from model-centric development toward data-centric development—“when you systematically change or enhance your datasets to improve the performance of the model” (reference 12).
An example of how to improve data quality is to create descriptive labeling guidelines to mitigate recall bias when using data labeling companies like AWS’ Mechanical Turk. Additionally, responsible AI frameworks should be in place to ensure data governance, security and privacy, fairness, and inclusiveness.
Data privacy through federated learning. The importance of data privacy has not only forged the path to new laws (e.g., GDPR and CCPA), but also new technologies. Federated learning enables ML models to be trained using decentralized datasets without exchanging the training data. Personal data remains in local sites, reducing the possibility of personal data breaches.
Additionally, the raw data does not need to be transferred, which helps make predictions in real-time. For example “Google uses federated learning to improve on-device machine learning models like ‘Hey Google’ in Google Assistant, which allows users to issue voice commands” (reference 13).
AI in smart factories
Maintenance, demand forecasting, and quality control are processes that can be optimized through the use of artificial intelligence. To achieve these use cases, data is ingested from smart interconnected devices and/or systems such as SCADA, MES, ERP, QMS, and CMMS. This data is brought into machine learning algorithms on the cloud or on the edge to deliver actionable insights. According to IoT Analytics (reference 14), the top AI applications are:
predictive maintenance (22.2 percent)
quality inspection and assurance (19.7 percent)
manufacturing process optimization (13 percent)
supply chain optimization (11.5 percent)
AI-driven cybersecurity and privacy (6.6 percent)
automated physical security (6.5 percent)
resource optimization (4.8 percent)
autonomous resource exploration (3.8 percent)
automated data management (2.9 percent)
AI-driven research and development (2.1 percent)
smart assistant (1.6 percent)
other (5.2 percent).
Vision-based AI systems and robotics have helped develop automated inspection solutions for machines. These automated systems have not only been proven to save human lives but have radically reduced inspection times. There have been significant examples where AI has outperformed humans, and it is a safe bet to conclude that several AI applications enable humans to make informed and quick decisions (reference 15).
Given the myriad additional AI applications in manufacturing, we cannot cover them all. But a good example to delve deeper into is predictive maintenance because it has such a large effect on the industry.
Generally, maintenance follows one of four approaches: reactive, or fix what is broken; planned, or scheduled maintenance activities; proactive, or defect elimination to improve performance; and predictive, which uses advanced analytics and sensing data to predict machine reliability.
Predictive maintenance can help flag anomalies, anticipate remaining useful life, and provide mitigations or maintenance (reference 17). Compared to the simple corrective or condition-based nature of the first three maintenance approaches, predictive maintenance is preventive and takes into account more complex, dynamic patterns. It can also adapt its predictions over time as the environment changes. Once accurate failure models are built, companies can build mathematical models to reduce costs and choose the best maintenance schedules based on production timelines, team bandwidth, replacement piece availability, and other factors.
Bombardier, an aircraft manufacturer, has adopted AI techniques to predict the demand for its aircraft parts based on input features (i.e., flight activity ) to optimize its inventory management (reference 18).
This example and others show how advances in AI depend on advances associated with other Industry 4.0 technologies, including cloud and edge computing, advanced sensing and data gathering, and wired and wireless networking.
(Courtesy of International Society of Automation)
About The Authors:
Ines Mechkane is the AI Technical committee chair of ISA’s SMIIoT Division. She is also a senior technical consultant with IBM. She has a background in petroleum engineering and international experience in artificial intelligence, product management, and project management. Passionate about making a difference through AI, Mechkane takes pride in her ability to bridge the gap between the technical and business worlds.
Manav Mehra is a data scientist with the Intelligent Connected Operations team at IBM Canada focusing on researching and developing machine learning models. He has a master’s degree in mathematics and computer science from the University of Waterloo, Canada, where he worked on a novel AI-based time-series challenge to prevent people from drowning in swimming pools.
Adissa Laurent is AI delivery lead within LGS, an IBM company. Her team maintains AI solutions running in production. For many years, Laurent has been building AI solutions for the retail, transport, and banking industries. Her areas of expertise are time series prediction, computer vision, and MLOps.
Eric Ross is a senior technical product manager at ODAIA. After spending five years working internationally in the oil and gas industry, Ross completed his master of management in artificial intelligence. Ross then joined the life sciences industry to own the product development of a customer data platform infused with AI and BI.
Instrumentation Basics 