Embedding is a mapping between anything and a vector.
Each word is represented as a vector.
Similarity between words can be measured in the vector space called embedding space.
Given a center word (with one-hot coding input), Word2Vec predicts context words (words before and after center word).
Because we are using one-hot vector, a column of weight becomes the hidden layer - called word vector.
Transport layer provide logical communication between application processes running on different hosts.
Sender breaks application messages into segments, then passes to network layer.
Receiver reassembles segments into messages, then passes to application layer.
Transport layer is communication between processes. (TCP, UDP)
Network layer is communication between hosts. (IP, Router)
CDN was first introduced to distribute non-video content, but it is widely used in video streaming these days.
Video traffic is a major consumer of Internet bandwidth. (80% of residental ISP traffic in 2020!)
Video is sequence of images displayed at constant rate. Because sending the video as is would eat up too much traffic, we use coding!
A single frame is chosen and called the keyframe - other frames are encoded as differences from the keyframe.
Spatial coding: Remove redundancy within image.
Temporal coding: Remove redundancy between image. (i.e. between frame i and frame i+1)
CBR (constant bit rate): Video encoding rate is fixed.
VBR (variable bit rate): Video encoding rate changes as amount of spatial, temporal coding changes.
Self scalability - new peers bring new service capacity, and new service demands!
How much time does it take to distribute file (size F) from one server to N peers?
We need to test our ideas!
A prototype is a simplified and incomplete models of a design used to:
Rotation is very nonintuitive;
In geometry, rotation is a movement that keeps at least one point fixed.
But in physics, rotation is extremely complex!
Intermediate axis theorem: Rotation around major and minor axis is stable, but rotation around intermediate axis is unstable!
The rotation axis keeps chaning, and the object can even be flipped!
This is actually similar to Retinal Ganglion Cell (RGC, 망막 신경절 세포)
RGC is connected to a group of photoreceptors which detects the existence of light.
The pattern and motion information is then transferred to Primary Visual Cortex (V1).
V1 neuron receives information from multiple RGCs and output single response, just like weighted sum + activation function.
V1 is known to have 6 layers which is similar to filters in CNN.
V1 neurons in the same column (ocular-dominance column) detects line segment.
Adjacent columns detects slightly different line orientations.
A simple cells detects a line segment of a specific direction.
A complex cell receives information from multiple simple cells, and detects a line segment.
This is similar to layers - it can extract high level feature from low level feature!
Internet hosts (or routers) has IP address and name.
DNS maps between IP address and name and vice versa.
DNS is a distributed database implemented in hierarchy of many name servers.
It is actually an application-layer protocol: hosts and DNS servers communicate to resolve names. (address/name translation)
DNS is at network's edge; It's not core function of network!
DNS uses TCP, UDP, or even TLS!
E-mail has three major components:
User Agents compose, edit, read mail messages.
Using data collected in the analysis phase,
Elastic user can mean anyone and thus no one.
Vauge audience leads to unfocussed design.
Trying to design for everyone often means designing for no one!