Shaona Ghosh (Postdoc at University of Cambridge, Engineering) talks about ways of leveraging data in order to build AI and machine learning algorithms that can learn and make intelligent predictions about the product.
Data driven AI/Machine Learning algorithms
Machine learning algorithms are bottom up AI techniques that are extremely good in finding patterns in data, learning from the patterns in existing data and predicting on new data. In the first part of her talk, Shaona discusses the machine learning tools that are available online and what can be done if there are no such off-the-shelf AI algorithm that suits the needs of the product.
A very important step that is quite often ignored is that of making the data ready for AI algorithms. There should be reasonable clarity on how the data in consideration is sampled from the larger dataset, followed by a pre-processing step that cleans, scales and normalizes the data. Reasonable thought is required on what the data records represent. Preliminary visualization and analysis can reveal a lot more information about your data that would have been otherwise unknown. Assumptions based on domain knowledge of the data can accentuate this stage and lead to better ways of visualizing or representing the data for the next step where machine learning algorithms work on them.
Deep Learning Algorithms
In the next part of her talk, Shaona discusses a powerful machine learning framework called deep learning. Following up on her pointers of how one can represent data by hand tuning them, a deep neural network automatically represents the data for you without the need of human intervention. Such data representation is often superior to human perception. The result is more accurate pattern recognition and prediction results, even surpassing human accuracy for certain problems.
The key takeaway from her talk is that not every machine learning algorithm is applicable for every realm of data; it’s important to know where deep learning can be used for what kind of data, where to look for software and hardware resources for deep learning, what are the types of deep learning algorithms and who are the best minds working on this. Her talk makes an attempt to cover as many such points as possible in order to nudge the audience in the right direction.
Comments 0