The neuro-network paradigm is so successful. AI researchers are looking at the following;
1. How to train AI on Small Data rather than Big Data.
2. How to accelerate learning of AI.(e.g. Master AI and Student AI working in tandem)
3. How to go inside the black box, to the point we don't know, to understand how AI works past a certain point.
4. How to enable AI train in one domain to carry the learning to another domain.
5. Integrated learning in AI
6. How to create robots that can navigate a room of people, understand what is going on, and be of assistance.
And my favorite...
Human-AI integration
Enabling the human brain to connect to the external world of machine learning, to adapt enhanced cognitive abilities. This is the merging of human intelligence and AI intelligence. Within a decade, tasks will be done differently, as human-AI collaboration grows.
HOW GOOGLE USES AI TO ENHANCE USER EXPERIENCE
SEARCH
1. Voice Search
2. Rank brain- neuro networks to increase relevance
3. Multitask unified model- to integrate content and context at a level not achieved so far. For example is one needs to know the best school for their child, Google will deploy the knowledge of the specific user to give them the best answer according to their needs and means.
Smartphone AI
Google’s smartphones have AI abilities in the device, not just in the cloud. There are also camera sensors that guides color balancing, brightness, and exposure, especially for people of darker skin.
Biomedical companies are also deploying AI to recreate DNA. Other ways that AI is used in the medical industry include- employing machine learning to transcribe medical documents, remotely treating patients, and generating end to end drug discovery/ development.
Tips while creating an AI model.
1. Be very specific with the data set employed
2. Look for missing data
3. Assess features
4. Assess the size of the database
5. Look for missing data
Two things needed: 1. Machine Learning Algorithm 2. A Data Set.
Netflix's presentation of recommendation is a perfect example of successful unsupervised machine learning. The algorithm accesses the data base of your historical watches and compares these to other users whose watching behaviors are similar to yours.
How to train and AI model
Assemble a detailed data base with the input and output variable
Clean up the data
Employ a supervised learning algorithm to train an ML model on each algorithm and pick an algorithm that beats others in the model
Specify parameters to guide the algorithm in carrying out the task.
State how the model should be split, the number of epochs., the number of iterations, and number of hidden layers.
Comentarios