The introduction and use of Artificial Intelligence in Imago have now taken on a key and increasingly essential role, thanks to the systematic application of the most recent and innovative Machine Learning and Anomaly detection algorithms to traditional vision models.
But how and when was Artificial Intelligence born? What are the pros and cons of using it? And what differences exist between the different Machine Learning, Deep Learning and Neural Network technologies? With great skill in explaining the most complicated issues with maximum clarity, Matteo Baldelli and Andrea Tiberti introduce you to the origins of this discipline and the operating principles of the Artificial Intelligence technologies that Imago has already successfully integrated into the software of its vision systems.
Eng. Matteo Baldelli: He graduated in Physics at the University of Pavia, with a specialist focus on image processing using Artificial Intelligence. Following extensive experimentation, he implemented the Imago algorithms in applications of all levels of complexity, leading to a consolidated use of the main AI technologies.
Eng. Andrea Tiberti: He obtained a master's degree in Computer Engineering from the University of Brescia in 2019. During his academic career he attended many courses dedicated to Machine Learning and Artificial Intelligence. Motivated by a great passion for the topic of AI, he therefore applied his profound skills to the applications and research of Imago projects.
Interview
1) “Artificial Intelligence” (AI) has been named word of the year 2023 by the Collins Dictionary, the authoritative English dictionary published by the British publishing house of the same name. The choice reflects the enormous impression this technology is having on society. So what is artificial intelligence? How and when was it born?
Matteo Baldelli:
The naming of "Artificial Intelligence" as the Collins Dictionary's Word of the Year 2023 reflects the growing importance of this technology in contemporary society and its widening diffusion across a wide range of sectors. The idea of creating machines that can simulate human intelligence dates back several centuries, but the modern concept of artificial intelligence was born in the post-World War II period. In 1956, the Dartmouth Conference took place at Dartmouth College, which is generally considered the official beginning of the field of artificial intelligence. The scholars present at this conference set themselves the goal of developing computer programs capable of simulating human intelligence. In the years since, several techniques and approaches have been developed to achieve artificial intelligence, including artificial neural networks, decision trees, genetic algorithms, and much more. However, AI has gone through several phases of ups and downs over the decades, with periods of great enthusiasm followed by disappointments (such as the so-called “AI winter”) due to technical limitations and unmet expectations. In recent years, thanks to increases in computing power, access to large amounts of data, and developments in machine learning and deep neural networks, artificial intelligence has made enormous progress. Today, AI is ubiquitous in many areas of daily life, from online shopping recommendations to autonomous vehicles, from medical diagnostics to financial data analysis, to name a few examples.
2) As we said, artificial intelligence is already part of our daily lives and is finding application in various sectors of industry. The world of vision also increasingly uses Machine Learning, Deep Learning and Neural Networks technologies. Can you give us a quick definition of these terms and their working principles?
Matteo Baldelli:
Machine Learning is a branch of artificial intelligence that deals with developing algorithms and models that allow computers to learn from data and improve their performance over time without being explicitly programmed to do so. Machine learning algorithms analyze data to identify patterns and relationships, and use this information to make predictions or decisions. Deep Learning is a subcategory of Machine Learning that uses artificial neural networks composed of several hidden layers (hence the term “deep”) to learn hierarchical representations of data. Deep neural networks (Deep Learning) are in fact composed of layers of artificial neurons, inspired by biological neurons, each of which processes and transmits information through a set of weights that are adapted during the training process. These models are capable of automatically learning complex features and patterns in data, allowing them to achieve exceptional performance on tasks such as image recognition, natural language recognition, and more. Neural networks are therefore the algorithms that make up the deep learning category.
3) Compared to traditional vision systems, what are the pros and cons of using artificial intelligence?
Andrea Tiberti:
The use of artificial intelligence in the field of vision systems, if trained on a significant data set, allows the development of algorithms that are less sensitive to the variation of the parts analyzed and therefore guarantees a more robust and flexible analysis compared to traditional vision. The latter, however, require careful work in which it is necessary to define a greater number of parameters to be configured and modified in relation to the variation in environmental conditions, light and brightness of the part. Greater flexibility also translates into a reduction in the human intervention necessary to configure the parameters mentioned above. The main disadvantage in artificial intelligence algorithms is the fact that they are "black-box", that is, once trained, it is difficult to intervene to improve their performance or correct any unwanted behavior except through re-training them. On the other hand, traditional vision systems allow you to intervene precisely on the aspects you wish to modify or improve. I think the winning solution is, at the moment, a hybrid version capable of benefiting from the positive aspects of both approaches.
4) What are Imago solutions for AI in vision systems?
Andrea Tiberti:
The main state-of-the-art deep learning algorithms for image analysis have been integrated into our software. These algorithms guarantee even greater robustness against production variations and condition variations in general. We have also implemented anomaly detection algorithms starting from academic papers, which are trained using compliant production samples, without the need to acquire and analyze large quantities of defects, which are often difficult to find by customers.
5) Is this a complementary use or an increasingly systematic development?
Andrea Tiberti:
This is undoubtedly an increasingly systematic use, which has now become a standard in Imago. In recent years, in fact, the number of Imago vision systems with artificial intelligence algorithms on board has grown considerably, in step with our know-how on this topic. Our strength lies in the fact that we can combine these new algorithms with classic ones in order to compensate for any weaknesses and benefit from the strengths of both approaches.
6) What opportunities and scenarios do you foresee for the future of artificial intelligence at Imago?
Matteo Baldelli:
For about three years we have been constantly integrating the most important and current Deep Learning algorithms. The additions are often made starting from the reading of scientific papers. As is known, it is a constantly expanding sector in which "game-changing" technologies are released for our sector in a very short time. In addition to deep learning algorithms for image processing, we have also integrated machine learning technologies for process analysis, such as the analysis of time series of data extracted from our software. The goal is to continue on this path, moving at the same speed as a constantly evolving sector in order to keep Imago and its products always in step with the times.
7) How have these innovations helped Imago find solutions it might not have found otherwise, enabling its customers to overcome previously insurmountable challenges?
Matteo Baldelli and Andrea Tiberti:
These innovations have certainly allowed Imago to integrate new skills into its know-how to develop applications that would otherwise not be feasible. These are projects characterized by considerable variability in both production and environmental conditions. The results obtained represented a real leap forward for our partners, generating maximum satisfaction from their customers.
