Lecture 9 - Big Data and Artificial Intelligence

what are the definitions that encompass big data?

scale (magnitude of the data), diversity (the different types of multi modal data), complexity (how hard it is to intersect) and it requires new architecture and methods to manage and extract value from it

what are the 4 Vs of big data?

volume, variety, velocity and veracity

volume

the amount of data that is out there, seems to be increasing exponentially due to the mass generation of data production powered by social media

variety

the different formats, types and structures of data including text, numerical, video, static vs streaming data

• in order to extract knowledge all these types of data need to be linked together (a single app can do this since it generates and collects many types of data)

veracity

the rate at which data is being processed, data seems to be generating fast and by extension needs to be processed fast

velocity in E promotions

based on your current location, purchase history and what you like promotions will be sent right now for the store that is closest to you

velocity in healthcare monitoring

sensors monitoring your activities and body detects any abnormalities observed and those typically require immediate reaction/medical attention

veracity

how accurate our data is

• confidence tends to drop as data volumes go up causing uncertainty due to data inconsistency, its usually incomplete, there are ambiguities, latency and model approximations to maximize accuracy

statistical imputation

taking the mean of a large sum of data and trying to fill the missing pieces of data in a way that doesn’t affect average values for the group

what are the two main challenges in big data?

storage and analysis

analysis

connecting and integrating multi modal data, real time processing, integration into the cloud storage framework, need an intelligent system that can function in a similar manner to a human

what makes humans smart?

evolution, the fact that we are the dominant species, we often learn from our mistakes and we adapt ourselves in the environment to perform better

intelligence

the ability to acquire and apply knowledge and skills

AI

the capability of a machine to imitate intelligent human behavior, uses computers to model intelligent behavior with minimal human intervention, should be able to store, compute and learn

what is the alan turing test?

test where a human questioner is asked a series of questions to both respondents and after a specified time, the questioner tries to decide which terminal is operated by the human respondent and which terminal is operated by the computer

how do we model intelligent behaviour?

by looking at how the brain works

how does the brain function?

as a result of an immense number of neurons that fire signals

how do neurons fire signals?

they connect through synapses that propagate electrical impulses by releasing neurotransmitters

synaptic plasticity

the activity-dependent changes in the effectiveness of synapses or how synapses can alter the strength of their connections (this is what allows us the capacity to learn, the stronger the connection the more you learn)

what are the different types of machine learning

unsupervised learning, supervised learning and reinforcement learning

neural networks

computational equivalent of neurons

what is the point of unsupervised learning?

use to uncover naturally occurring patterns or groupings in data without targeting a specific outcome, determining pattern from unlabelled data by using the machine to cluster the similar data together

how is unsupervised learning used in healthcare?

goal of it is to uncover subsets of patients who share similar clinical or molecular characteristics and in theory are more likely to respond to targeted therapies directed at their shared underlying pathobiology

what is the role of supervised learning?

used to uncover the relationship between variables of interest and one or more target outcomes, these target outcomes must be know so you can ask the machine to model something that has predictive power

how is supervised learning used in healthcare?

if researchers wanted to know whether a set of clinical features like vtal signs per example, they could predict ICU mortality by applying this machine algorithm to a data set in which each patient record contains the set of clinical features of interest and a label specifying their outcome

reinforcement learning

learning done by trial and error where rewards and punishments are used as signals for positive and negative behavior and the goal is to find a suitable action model that would maximize the total cumulative reward of the agent

computer vision

the processing of an image to enable identification of image input and to provide an appropriate output combining reinforcement with image recognition and usually applied to video and image data