From the predictive text of our smartphones to the recommendation engines of our favorite shopping websites, machine learning (machine learning, ML) is already integrated into our daily routines. But machine learning is not stopped, but in constant evolution.
In recent years, it has progressed rapidly, largely thanks to improvements in data collection, processing power, and the development of more sophisticated algorithms.
Now, as we move into the second half of 2023, these technological advancements have paved the way for exciting new trends in machine learning. These trends not only reflect the continued advancement of machine learning technology, but also highlight its increasing accessibility and the increasingly crucial role of ethics in its applications.
From codeless machine learning to tinyML, these seven trends are worth observing in 2023.
Machine learning, or AutoML, is one of the most significant machine learning trends we are witnessing. Approximately 61% of decision makers in companies using AI claimed to have adopted autoML, and another 25% planned to implement it that year. This innovation is reshaping the process of creating ML models by automating some of its most complex aspects.
AutoML is not intended to eliminate the need for encoding, as is the case with ML platforms without code. Instead, AutoML focuses on automating tasks that often require a high level of experience and a significant investment of time. These tasks include preprocessing data, selecting features, and adjusting hyperparameters, to name a few.
In a typical machine learning project, these steps are performed manually by engineers or data scientists who have to iterate multiple times to optimize the model. However, AutoML can help automate these steps, thus saving time and effort and allowing employees to focus on solving higher-level problems.
Furthermore, AutoML can provide significant value to non-experts or those in the early stages of their ML journey. By eliminating some of the complexities associated with ML, AutoML allows these people to harness the power of machine learning without the need to thoroughly understand every intricate detail.
Tiny machine learning, commonly known as TinyML, is another significant trend that deserves our attention. The installation of tinyML devices is projected to increase from nearly 2 billion in 2022 to more than 11 billion in 2027. The engine of this trend is the power of tinyML to bring machine learning capabilities to small, energy-efficient devices, often called peripheral devices.
The idea behind TinyML is to run machine learning algorithms on devices with minimal computing resources, such as microcontrollers on small appliances, portable devices, and Internet of Things (IoT) devices. This represents a shift from cloud-based computing to local computing on the device, bringing benefits like speed, privacy, and lower power consumption.
It is also worth mentioning that TinyML opens opportunities for real-time decision making on the device. For example, a portable health tracker could take advantage of TinyML to analyze a user's vital signs and alert them to abnormal readings without the need to constantly communicate with the cloud, thus saving bandwidth and preserving privacy.
Generative AI has dominated headlines in 2023. Since the launch of OpenAI ChatGPT in November 2022, we have seen a wave of new AI technologies generating large technology companies such as Microsoft, Google, Adobe, Qualcomm, as well as countless innovations from companies of all sizes. These sophisticated models have opened up unprecedented possibilities in many fields, from art and design to data growth and beyond.
Generative AI, as a branch of machine learning, focuses on creating new content. It is like endowing an AI with a form of imagination. These algorithms, using various techniques, learn the underlying patterns of the data they train with and can generate new and original content that reflects those patterns.
Perhaps the best known form of generative AI is the adversarial generative network (GAN). GANs face two neural networks: one generator, which creates new data instances, and another discriminator, which tries to determine if the data is real or artificial. The generator continually improves its results to trick the discriminator, creating incredibly realistic synthetic data.
However, the field has expanded beyond the GAN. Other approaches, such as variational autocoders (VAE) and transformer-based models, have shown impressive results. For example, VAE is now used in fields such as drug discovery, where they generate new viable molecular structures. Transformer-based models, inspired by architectures like GPT-3 (now GPT-4), are being used to generate human-like text, allowing for more natural conversational AI experiences.
In 2023, one of the most notable advances in generative AI is the improvement and greater adoption of these models in creative fields. AI is now capable of composing music, generating unique works of art, and even writing compelling prose, broadening the horizons of creative expression.
However, along with its fascinating potential, the rapid advancement of generative AI carries significant challenges. As generative models are increasingly capable of producing realistic results, it is essential to ensure that these powerful tools are used responsibly and ethically. The possible misuse of this technology, such as the creation of deepfakes or other misleading content, is a major concern that will need to be addressed.
AI's interest and demand for technology, combined with the growing skills deficit in this field, has led more and more companies to opt for codeless machine learning solutions. These platforms are revolutionizing the field by making machine learning more accessible to a wider audience, including those with no programming or data science experience.
Codeless platforms are designed to allow users to create, train, and deploy machine learning models without writing any code. They often have intuitive and visual interfaces where users can manipulate preconstructed components and use established machine learning algorithms.
The power of the codeless ML lies in its ability to democratize machine learning. It opens the doors to business analysts, subject matter experts, and other professionals who know your data and the problems they need to solve, but who lack the programming knowledge that traditional machine learning often requires.
These platforms allow users to harness the predictive power of machine learning to generate ideas, make data-driven decisions, and even develop smart applications, all without the need to write or understand complex code.
However, it is crucial to note that while codeless ML platforms have done wonders to increase the accessibility of machine learning, they are not a complete substitute for understanding the principles of machine learning. While reducing the need for coding, interpreting the results, identifying and treating potential biases, and the ethical use of ML models still require a solid understanding of machine learning concepts.
Another crucial trend of machine learning in 2023 that should be highlighted is the growing focus on ethical and explainable machine learning. As machine learning models become more ubiquitous in our society, understanding how they make their decisions and ensuring that those decisions are made ethically has become paramount.
Explanable machine learning, often known as interpretable machine learning or explainable AI (XAI), is to develop models that make transparent and understandable predictions. Traditional machine learning models, especially the most complex ones such as deep neural networks, are often considered «black boxes » because their internal functioning is difficult to understand. The XAI intends that the decision-making process of these models is understandable to humans.
The growing interest in XAI is due to the need for responsibility and confidence in machine learning models. As these models are increasingly used to make decisions that directly affect people's lives, such as loan approval, medical diagnoses, or job applications, it is important that we understand how they make those decisions and that we can trust its precision and impartiality.
In addition to explainability, more and more attention is paid to the ethical use of machine learning. Ethical machine learning involves ensuring that models are used responsibly, that they are fair, impartial and that they respect the privacy of users. It also involves reflecting on the possible implications and consequences of these models, including how they could be misused.
In 2023, the boom in explainable and ethical machine learning reflects a greater awareness of the social implications of machine learning (as well as the rapid evolution of legislation regulating how machine learning is used). It is an acknowledgment that although machine learning has immense potential, it must be developed and used responsibly, transparently and ethically.
Another trend that is shaping the machine learning landscape is the increasing emphasis on machine learning operations or MLOps. According to a recent report, the global MLOps market is projected to grow from $ 842 million in 2021 to almost $ 13 billion in 2028.
In essence, MLOps is the intersection of machine learning, DevOps, and data engineering, with the goal of standardizing and streamlining the life cycle of developing and implementing machine learning models. MLOps' central objective is to bridge the gap between the development of machine learning models and their operation in production environments. This involves creating a robust process that enables fast, automated, and reproducible production of models, incorporating steps such as data collection, modeling, validation, deployment, monitoring, etc.
A significant aspect of MLOps is the focus on automation. By automating repetitive and slow ML life cycle tasks, MLOps can dramatically speed up time from model development to deployment. It also ensures consistency and reproducibility, reducing the chances of errors and discrepancies.
Another important facet of MLOps is supervision. It is not enough to deploy a model, but it is crucial to monitor its performance. MLOps encourages continuous monitoring of model metrics to ensure they work as expected and to quickly detect and address any performance deviations or degradation.
In 2023, the increasing emphasis on MLOps is a testament to the maturation of the field of machine learning. As organizations try to take advantage of machine learning at scale, efficient and effective operating processes are more crucial than ever. MLOps represents an important step forward on the road to operationalizing machine learning in a sustainable, scalable and reliable way.
The latest trend that is drawing attention in the field of machine learning in 2023 is multimodal machine learning. As the name suggests, multimodal machine learning refers to models that can process and interpret multiple types of data - such as text, images, audio, and video - in a single model.
Traditional machine learning models often focus on one type of data. For example, natural language processing models treat text, while conventional neural networks are excellent for image data. However, real-world data is often presented in several forms, and valuable information can be extracted when these different modalities are combined.
Multimodal machine learning models are designed to handle this diversity of data. They can receive different types of data, understand the relationships between them and generate exhaustive information that would not be possible with monomodal models.
For example, imagine a model trained in a movie dataset. A multimodal model could simultaneously analyze the dialogue (text), the expressions and actions of the actors (video) and the soundtrack (audio). This would provide a more nuanced understanding of the film compared to a model that analyzed only one type of data.
By 2023, we will see more and more applications taking advantage of multimodal machine learning. Multimodal learning is a trend that is redefining the possibilities in the field of machine learning, from more attractive virtual assistants capable of understanding speech and seeing images to health models capable of analyzing disparate data flows to detect cardiovascular diseases.
