Artificial intelligence (AI) is a broad field defined by the development of machines to perform tasks that would normally require human intelligence. A subset of AI is machine learning, which involves the development of computer algorithms that can learn from data to perform tasks without explicit instructions. Machine learning has risen to mainstream popularity in recent years with the advent of powerful AI tools such as ChatGPT, DALLE, and other generative AI tools. Additionally, you may be interacting with a machine learning algorithm daily without even recognizing it in applications such as driving, shopping, and even texting.
Types of Machine Learning Models
A machine learning algorithm, in essence, is a statistical model that performs mathematical operations on data to produce a desired output (prediction/inference). Machine learning algorithms can often be categorized into 3 major categories:
- Supervised Learning: These types of models utilize labeled (ground truth) data to train (aka teach) the model how to make predictions on unlabeled data where the labels represent the actual target outputs that you want the model to predict.
- Labeled data can come in various forms depending on the application, such as numerical or categorical values for data tables or segmentation masks for images. Additionally, labelled data is typically annotated manually or is collected from historical data.
- Prevalent examples include Gmail spam filtering, Goldman Sachs stock predictions for investing, and Apple FaceID.
- At Visikol, we utilize supervised learning for uses such as object detection for certain cell types or structures or classification of samples based on extracted data.
2. Unsupervised Learning: In contrast to supervised learning, unsupervised learning models do not require labeled data. Rather, these models take collected data as an input and try to find patterns and relationships within the data.
- A very prevalent example of this is seen in recommendation systems such as that of Amazon, Netflix, and YouTube.
- Visikol utilizes unsupervised learning mainly for bioinformatics with operations such as clustering of samples by their features or a principal component analysis to reduce multidimensional data.
3. Reinforcement Learning: On the other hand, reinforcement learning models utilize an agent to make decisions in an environment to reach a certain outcome. The agent is a mechanism that observes the current state of the environment it is in to take an action based on its set of rules (policy) where after taking an action, the agent is then rewarded and improves upon the policy based on the reward.
- Examples include Tesla’s self-driving cars, Google DeepMind to manage data center operations, and Ubisoft in NPC navigation in video games.
From finance to entertainment, machine learning has truly become an integral part of virtually every aspect of our lives whether we notice it or not. Machine learning is a broad field with many applications, which can be categorized into 3 major categories; supervised, unsupervised, and reinforcement learning. Supervised learning is used to make predictions and inferences on data whereas unsupervised learning is used to recognize patterns and relationships within datasets. Contrary to the other methods, reinforcement learning continuously operates to learn from its environment to reach a desired outcome.
At Visikol, we utilize machine learning regularly for research in drug discovery and clinical trials for our clients. Mainly, we utilize supervised and unsupervised learning to detect objects of interest and analyze data. If you require machine learning services for biomedical research, please feel free to contact us with any questions you may have.
Through this series of articles, it is my aspiration to introduce readers to machine learning and improve their understanding of what goes into creating machine learning models, so stay tuned for the next article!
Author
Jigar Patel
Senior Lead Image Analyst