Wednesday, July 20, 2016

Stanford CSP BUS 74 [Principles of Invention and Innovation], Session 4 Quiz 1

On July 19, 2016, Bloomberg Technology News reported that Google used its DeepMind AI technology to reduce power consumption in the company's data centers:
In recent months, the Alphabet Inc. unit put a DeepMind AI system to reduce power consumption by manipulating computer servers and related equipment like cooling systems. It uses a similar technique to DeepMind software that taught itself to play Atari video games, Hassabis said in an interview at a recent AI conference in New York.

The system cut power usage in the data centers by several percentage points, "which is a huge saving in terms of cost but, also, great for the environment," he said.

The savings translate into a 15 percent improvement in power usage efficiency, or PUE, Google said in a statement. PUE measures how much electricity Google uses for its computers, versus the supporting infrastructure like cooling systems.

Question 1. Using the system model, name the functional element that DeepMind technology helps to improve directly.
Question 2. Based on what you know about the improved element, describe other functional elements within the same system.


3 comments:

Hardik Nagrecha said...

[Hardik Nagrecha] To get the discussion started, here are a few pointers I have listed down. (Please note that these pointers represent my personal opinions on the subject matter and conclusions drawn after reading the articles and links mentioned below)

Question 1. Using the system model, name the functional element that DeepMind technology helps to improve directly.

DeepMind technology perfectly fits in the control segment of the system model. Their technology learns the consumption of electricity by the servers and then regulates usage based on specific parameters (such as server demand, etc).

As mentioned in the article on the matter, (http://www.theverge.com/2016/7/21/12246258/google-deepmind-ai-data-center-cooling) DeepMind's technology is a general framework for deep reinforcement learning that is replacing existing control techniques. The comparison plot when the deep learning based control is used and not used reaffirms the conclusion made earlier (shown in the reference link).

Question 2. Based on what you know about the improved element, describe other functional elements within the same system.

Looking at the problem using the system model, the varied parts are as follows:

Source: Electricity from the power grid
Distribution: Internal wiring for distributing the electricity (packaged payload) throughout the data center
Tool: Servers supporting the Google infrastructure
Control: Software to regulated consumption of electricity at the data center.

The reason why DeepMind’s framework could be applied to this problem was because Google had existing knowledge of the distribution and consumption of electricity of the data centers along with performance analysis for the old control techniques. The new control technique worked well because existing mathematical models failed to encompass all possible heuristics to control power consumption by the servers.

It is clear from the article that Google did install more sensors in their data centers to get better data on temperatures. This is the sensing part of the control subsystem (not evident from the system model) which is crucial since the software does need to know the state of the system to exercise effective control.

Leonardo Vieira said...

Question 1.

The functional element that DeepMind technology helps to improve directly are the computer servers and it's related equipment.
The system model in this case would be presented as the following:

Source: Power sources
Distribution: Energy Cords
Tool: Computer servers and it's related equipment
Package Payload(s): Energy
Control: DeepMind AI technology


Question 2.
I am not sure if I understood this question quite right, but other functional elements within this system would be:
-A regular Computer (which would be controled by a human)
-Air conditioner (responsible for the cooling process, which would also be controled by a human)

Brian said...

The system elements of the Google / DeepMind platform:

Source: Power (energy) used to run Google's data centers
Distribution: Power grid
Tool: Data center
Package Payload(s): Energy used to power Google's data centers
Control: DeepMind AI that helps to regulate and control the energy required to run Google's data centers.

Other functional elements within the system: data / information stored within the data center, servers within the data center, computers and other devices that leverage the data center for information, other platforms and systems using the data center.

- Brian Favat