AI, Machine Learning & Deep Learning
Lot of talk is prevailing aound these days around deep learning.
Find below a working prototype example, use case of deep learning:
Use image or take picture of a thing around you, to find what are similar things available in your favorite retail shop
Here are my quick thoughts on deep learning
- Artificial Intelligence (AI) is not new
- Remember Lisp, fixed rule based programming and case expert systems
- Is it going to remain popular and soon going to create thinking solutions replacing humans ?
- In the past also there’s been AI Hype and it fizzled down
- This time around it has already reached near human level accuracy for many perceptual tasks i.e. image , voice and text processing
- Deep Learning should not be regarded as how the human brain works
- Popularly misnomerally regarded as neural networks computing however there’s no evidence of human brain works based on deep learning data computing.
- It is not deeper learning, requires high school or college level mathematical computing understanding only
- Statistical and probalistic computating algrothims models are the foundations of Machine Learning.
- New Rules could be created ( No hard coded relationships)
- Results could be explained by mathematical model responsible
- Is Machine Learning actually shallow learning, perhaps yes
- Deep learning is simply Layered Computing
- Series of Data Transformations by multiple layers, supervised machine learning had already introduced loss and optimzer functions to train the models and deep learining uses the same for backpropgation of signals to adjust the weights for different computing layers
- Not easy to explain the results as in machine learning algorithms
- Layers do Tensor Computing mathematical operations
- Tensors are multi dimensional vectors or arrays
- Multi Core CPU’s are great , but GPU computing is needed for Deep learning
- Deep Learning can be used for Classification Problems to predict the binomial, multi category or continuos value outcome
- Unlike Machine Learning , heavy lifting of feature engineering is done by computational layers
- Pre-processing of input data , specially vectorization is still needed
- Need to Make Meaningful training data along with validation and test data , which may mean feature engineering is still needed
To me, discipline of Deep learning is sub set of Machine Learning , which in turn is subset of Artificial Intelligence
Google Cloud has already pre-trained models for image classification. These are trained models using googles Tensor Flow computation library.
Developers can easily benefit from these google cloud API’s and need not know to the extent of Data Scientist to make good use of it
With Enterprise systems like SAP providing techologies and application platform using HANA accessing enterprise data and machine learning and using deep learning compute engines like the ones provided by google are much much easier than it was in initial days of Artificial Intelligence computing, so seems a promising future is ahead, for not to overlook it, hence my experimentation and results are shared below