what is the best way to leaen AI Coding
Answers
Answered by
2
Use an API that offers developers the ability to build end-user applications. This enables you to skip through all the theory and actually build something useful in a few days rather than spending years learning Bayesian and other ML techniques. Here's a signup for our new API:Nathan Awaits... - AI for everyone.This is based on a new form of neural network that we have been developing for 10+ years: SDKs for building machine learning applications. If you don't like ours, you might want to tryGoogle Prediction API, Diffbot,MALLET or any of the open-source tools. There are many dozen forms of AI -- most of which require a lot of specialized training to use.
Answered by
2
Short answer: Use an API that offers developers the ability to build end-user applications. This enables you to skip through all the theory and actually build something useful in a few days rather than spending years learning Bayesian and other ML techniques. Here's a signup for our new API: Nathan Awaits... - AI for everyone. This is based on a new form of neural network that we have been developing for 10+ years: SDKs for building machine learning applications. If you don't like ours, you might want to try Google Prediction API, Diffbot, MALLET or any of the open-source tools. There are many dozen forms of AI -- most of which require a lot of specialized training to use.
Long answer: Each form of AI has strengths and weaknesses. You need to decide where you want to spend your time.
First you must determine what type of problems you want to solve. What is your use case, inputs and outputs?
Use Cases
Getting to "click click cool" is usually a long road - even for a really skilled team. As a beginner, I suggest you go with something super simple -- such as document similarity analysis. For example, mining a large corpus of documents to find contextual similarities. We recently built a proof-of-concept (POC) for a customer to compare 20,000 legal documents to find those that contain provisions that contain non-complete clauses. Traditional keyword searches fail in this situation since the concept of "non-compete" can be expressed in many different ways using the same words that also described other ideas. It took one of our engineers about a month to build a POC that became a commercial product within 3 months. There are many opportunities to build similar applications for different vertical markets -- such as medicine (using PubMed), insurance (fraud detection and claims processing), matching job candidates to positions (using LinkedIn), etc.
Inputs
The choice of the machine learning portion of AI technology will be driven by the type of data (inputs) you desire. There are three basic classes of data:
Unstructured -- such as human languages. These are data sets that do not fit into tables. They are most often managed with NoSQL databases.
Structured -- such as time-correlated signals, values, etc. These are datasets that easily fit into tables. They are most often managed with relational databases.
Visual -- such as images and video. These are data sets that are usually managed by storing entire files and organized using manually input tags.
Each type of data has its own unique semantic grammar. Generally speaking, there are many choices for structured data -- from libraries in MATlab and R to commercial products such as Numenta Grok or one of the many open-source libraries such as Machine Learning dlib. Unstructured data tends to be more challenging and dominated by tools such as LatentDirichletAllocation (LingPipe API).
Outputs
Sex sells. In my experience, making AI look cool is where many applications fail. If you only have 3 months, simplify the problem to the point where you can do a very easy layout. I suggest getting a UI/UX partner for the project. We find that Gephi is an excellent open source library for graphs that you can build really great visualizations in very little time.
Long answer: Each form of AI has strengths and weaknesses. You need to decide where you want to spend your time.
First you must determine what type of problems you want to solve. What is your use case, inputs and outputs?
Use Cases
Getting to "click click cool" is usually a long road - even for a really skilled team. As a beginner, I suggest you go with something super simple -- such as document similarity analysis. For example, mining a large corpus of documents to find contextual similarities. We recently built a proof-of-concept (POC) for a customer to compare 20,000 legal documents to find those that contain provisions that contain non-complete clauses. Traditional keyword searches fail in this situation since the concept of "non-compete" can be expressed in many different ways using the same words that also described other ideas. It took one of our engineers about a month to build a POC that became a commercial product within 3 months. There are many opportunities to build similar applications for different vertical markets -- such as medicine (using PubMed), insurance (fraud detection and claims processing), matching job candidates to positions (using LinkedIn), etc.
Inputs
The choice of the machine learning portion of AI technology will be driven by the type of data (inputs) you desire. There are three basic classes of data:
Unstructured -- such as human languages. These are data sets that do not fit into tables. They are most often managed with NoSQL databases.
Structured -- such as time-correlated signals, values, etc. These are datasets that easily fit into tables. They are most often managed with relational databases.
Visual -- such as images and video. These are data sets that are usually managed by storing entire files and organized using manually input tags.
Each type of data has its own unique semantic grammar. Generally speaking, there are many choices for structured data -- from libraries in MATlab and R to commercial products such as Numenta Grok or one of the many open-source libraries such as Machine Learning dlib. Unstructured data tends to be more challenging and dominated by tools such as LatentDirichletAllocation (LingPipe API).
Outputs
Sex sells. In my experience, making AI look cool is where many applications fail. If you only have 3 months, simplify the problem to the point where you can do a very easy layout. I suggest getting a UI/UX partner for the project. We find that Gephi is an excellent open source library for graphs that you can build really great visualizations in very little time.
Similar questions