“Artificial intelligence will reach human levels by around 2029. Follow that out further to, say, 2045, and we will have multiplied the intelligence – the human biological machine intelligence of our civilization – a billion-fold.”
– Ray Kurzweil, American inventor and futurist.
As we know, “Data” is the new power, and companies around the globe are trying to leverage this power in their businesses. Whether that business is:
- Healthcare: As the biomedical data is increasing, AI can provide a wide variety of services to aid humans. It can be diagnosis issues, drug invention, virtual healthcare, etc.
- Cybersecurity: AI tools are being used by cybersecurity companies to detect viruses and malware. AI systems are trained to identify even the smallest behavior of malware attacks.
- Recommending movies or products: Companies are using the recommendation systems to recommend movies or products to their users—for example, Netflix, Amazon. Recommendation systems are built by a technique called matrix factorization.
- Business analysis: Companies are using the power of AI for automation of the business process, collecting observation through data analysis, and using virtual support to engage with the customers and employees.
- Banking sectors: In banking sectors, AI is being used extensively for detecting fraudulent transactions, automation of the banking processes such as virtual support for account info, loan offer customization based on the user’s expenditures.
- Facial recognition: AI uses the facial features mapping to match the faces which are stored in the database. If the face exists in the database, then the match is true otherwise false.
People are experimenting with Machine Learning to get ahead… or simply keep up.
What are they doing exactly?
Top 5 trends in machine learning that you should look out for in 2020 and 2021
According to a report by McKinsey, 50% of the population of the USA suffers from a chronic disease, and 80% of medical care fees are spent on treatments.
Let’s see, in what significant healthcare sectors AI is being used extensively.
Top examples of AI in healthcare
Cancer Diagnosis with AI: Pathologists are using AI in healthcare to make a more accurate diagnosis. The goal can be achieved by collecting the data for different types of cancers like Histopathologic Cancer, Breast Cancer, Cervical Cancer, and further, using this data for making a predictive model. For example, PathAI, which is helping patients to diagnose the disease more accurately.
Developing New Medicines with AI: Biopharmaceutical companies are facing the challenges of overcoming the high attrition rates in drug development.
The Biopharmaceutical industry is collaborating with the AI industries to overcome these challenges.
For example, Atomwise is the first Deep Learning technology for novel small molecule discovery. It has assisted in the invention of new potential medicines for 27 diseases, and it is working with top institutes like Harvard University and Stanford University, as well as Biopharmaceutical companies.
Streamline Healthcare with AI: The number of patients around the globe is increasing every day. To process all the information about every patient, we need automated methods and systems. AI is helping healthcare facilities for better management of patient’s data.
For example, OLIVE is a platform designed to automate the healthcare industry’s tasks.
The world is going digital, with that, so is our money.
The total transactional values in the digital payments segment will be 4,406,431 million dollars in 2020, and it will be 8,266,917 million dollars by 2024.
All these transactions will be stored and processed efficiently. We can use these transactional data to improve our financial industry with the help of AI in 2020-21.
For example, Dataminr pulls information from various text sources and provides the user with a timeline in which each important event takes place. Dataminr uses NLP because it allows for analyzing text data.
Top examples of AI in Finance
Trading: Machine learning algorithms can be used to conduct trades autonomously. We can take the help of attributes like price, volume, time, but also tweet sentiment or weather data to make a machine learning system that beats the market (I am not saying it is easy :)). The algorithm can learn and adapt to real-time changes for making more accurate predictions. For example, Kayrros is a data analytics company that helps market participants to make better investment detection.
Fraud Detection: Using digital payments comes with many risks as well. In 2018, 24.26 Billion dollars were lost due to fraudulent transactions worldwide. Machine learning is ideally suited to fight deceitful financial factors effectively. British company AimBrain uses deep learning to target new account fraud and account takeover threats.
The model could be used to train the data while labeling every transaction if it was a fraud or not. Then we can use metrics like precision and recall to make a model suit our risk profile, adjusting to our costs of false positive and false negative predictions.
Banking: Banks are using machine learning for customer services, investment modeling, risk prediction, risk prevention, and investments. For example, we can provide personalized offers based on the user’s financial behaviors. So, if a client is looking for a house, then it might be useful to make a customized house loan offer for that user. Envestnet is a data aggregator and analysis company which helps in banking sectors extensively.
3. GAN (General Adversarial Networks)
Generative Adversarial Networks, or GAN for short, is an approach to generative modeling using deep learning methods, such as CNN.
GAN involves using a model to generate new, similar-looking data based from the data our network was trained on, such as images.
GANs can be used to generate the image datasets, human faces, cartoon characters, translate text-to-image, translate image-to-text, 3D object generation, etc. There are countless use-cases for GANs, but not all of them are good for society.
The most recent groundbreaking breakthrough in the GAN applications is deepfake.
Tero Karras, in their 2017 paper titles “Progressive Growing of GANs for Improved Quality, Stability, and Variation.” demonstrated the generation of realistic images of human faces.
The model has trained on celebrities’ faces, which means that the resultant faces will be having an extension of the existing personalities.
But don’t get fooled by the miracles it is performing with the data. The social impact that generating Deepfake images can have, is something we need to deal with soon. As it stands today, anyone’s reputation can be ruined in a matter of seconds with techniques available in publicly accessible repositories.
4. Reinforcement Learning
“Reinforcement learning (RL) is an area of machine learning concerned with how software agents ought to take actions in an environment to maximize the reward.
Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning.” – Wikipedia
WANT TO LEARN MORE ABOUT RL?
How to Make Sense of the Reinforcement Learning Agents? What and Why I Log During Training and Debug
RL is very appealing because it feels like the learning we observe every day. Let me give you an example.
Say you have a newly adopted dog who is only a month old. To teach your dog what is good and bad, you will use a reward system.
If the dog obeys you, you give him a cookie, and if the dog doesn’t, you scold him.
That is how you use positive reinforcement to train your dog for what you consider is the correct behavior.
The AI agents never received explicit instructions about how to play; they learned all by themselves.
After millions of simulations, AI agents learned to manipulate their environment:
- Hiders, for example, learned to build small forts and barricades.
- Seekers, in response, learned how to use ramps to scale the walls and find the hiders.
Isn’t it amazing?
The OpenAI techniques can be extrapolated to other AI scenarios as well by using the potential of multi-agent competitive environments as an influencer for learning without using any supervisor.
5. AR/VR (Augmented/Virtual Reality)
Augmented reality bridges the gap between virtual and physical reality.
The visual data that AR applications collect can be used in Image Sensing.
Augmented reality and AI are separate but complementary technologies. They both can leverage each other in order to build something magnificent.
Top examples of AR/VR with Machine Learning
Image or Scene Labeling: AI models are built by using the frames of a camera, which can help in classifying the location by labeling them where each frame is considered as an individual image. This is a great article here about image labeling.
Object Detection: Camera frames passed to an AI model that can estimate the position and size of objects within a scene. Location information can be further used to form a box around that object. For example, AnyVision can help in identifying a person or an object, even in the large crowds.
Pose Estimation: It is defined as the problem of the localization of human joints in images or videos.
There are two types of pose estimation:
- 2D Pose Estimation – Calculates the coordinates (x, y) for each joint from an RGB image.
- 3D Pose Estimation – Calculates the coordinates (x, y, z) for each joint from an RGB image.
Pose Estimation is heavily used in Action Recognition, Animation, Gaming, etc. This is a great blog to deepen your knowledge for human pose estimation.
We have seen five major sectors where AI is being used extensively. But AI is not limited to only Finance, Healthcare, GANs, AR/VR, and Reinforcement Learning.
It’s a technique which can be used anywhere where the data is being accumulated in large amounts. The data can describe stocks, planetary objects, or even human DNA.
Machine learning is applicable everywhere. The idea that machines could think and perform tasks, just like we humans do, is not that far 🙂
ML Experiment Tracking: What It Is, Why It Matters, and How to Implement It
10 mins read | Jakub Czakon | Posted November 26, 2020
Let me share a story that I’ve heard too many times.
”… We were developing an ML model with my team, we ran a lot of experiments and got promising results…
…unfortunately, we couldn’t tell exactly what performed best because we forgot to save some model parameters and dataset versions…
…after a few weeks, we weren’t even sure what we have actually tried and we needed to re-run pretty much everything”
– unfortunate ML researcher.
And the truth is, when you develop ML models you will run a lot of experiments.
Those experiments may:
- use different models and model hyperparameters
- use different training or evaluation data,
- run different code (including this small change that you wanted to test quickly)
- run the same code in a different environment (not knowing which PyTorch or Tensorflow version was installed)
And as a result, they can produce completely different evaluation metrics.
Keeping track of all that information can very quickly become really hard. Especially if you want to organize and compare those experiments and feel confident that you know which setup produced the best result.
This is where ML experiment tracking comes in.Continue reading ->