Quartz sync: Jul 19, 2024, 5:36 AM

README.md

@@ -1,18 +1,5 @@
-# Quartz v4
+## My Digital Garden
 
-> “[One] who works with the door open gets all kinds of interruptions, but [they] also occasionally gets clues as to what the world is and what might be important.” — Richard Hamming
+This is my [_Digital Garden_](https://maggieappleton.com/garden-history), a working collection of notes on AI🧠 and Software Design✨.
 
-Quartz is a set of tools that helps you publish your [digital garden](https://jzhao.xyz/posts/networked-thought) and notes as a website for free.
-Quartz v4 features a from-the-ground rewrite focusing on end-user extensibility and ease-of-use.
-
-🔗 Read the documentation and get started: https://quartz.jzhao.xyz/
-
-[Join the Discord Community](https://discord.gg/cRFFHYye7t)
-
-## Sponsors
-
-<p align="center">
-  <a href="https://github.com/sponsors/jackyzha0">
-    <img src="https://cdn.jsdelivr.net/gh/jackyzha0/jackyzha0/sponsorkit/sponsors.svg" />
-  </a>
-</p>
+Built using Obsidian & Quartz.

content/artificial intelligence/Artificial Intelligence.md

@@ -0,0 +1,10 @@
+> See [[Intelligence]]
+
+### Motivation
+
+1. automation of repeated physical solutions
+   - [[Industrial revolution]] (1750 - 1850) [[Machine Age]] (1870 - 1940)
+2. automation of repeated mental solutions
+   - [[Digital revolution]] (1950 - now) and [[Information Age]]
+3. allow machines to find solutions themselves
+   - We then only needs to specify a problem and/or goal

content/artificial intelligence/Intelligence.md

@@ -0,0 +1,22 @@
+## Intelligent Agent
+an **agent** acting in an **intelligent** manner. It perceives its **environment** through **observations**, takes actions **autonomously** in order to achieve **goals**, and may improve its performance with **learning** or acquiring knowledge. 
+
+## Intelligence
+To be able to learn to make decisions to achieve goals. How efficiently an **intelligent agent** is able to achieve its goals
+## Environment
+world in which the **agent** operates
+#### Environment State
+complete description of the state of the **environment**
+#### Observation
+partial description of the state of the **environment**
+#### Full Observability
+Observation = Environment state
+
+## Goals
+
+#### Terminal Goals
+The ultimate goal of an agent
+#### Intermittent Goals
+A necessary goal to achieve a **terminal** goal
+#### Utility Function/Rewards
+A function that returns how good or bad the agent is doing, how well it is succeeding at achieving its goals

content/artificial intelligence/cognitive robotics/Cognitive Robotics.md

@@ -0,0 +1,28 @@
+Field that combines insights and methods from [[Artificial Intelligence]], as well as **cognitive and biological sciences**, to [[Robotics]]
+
+> **What is Cognition Anyway?**
+Cognition is the process by which an autonomous system perceives its environment, learns from experience, anticipates the outcome of events, acts to pursue goals, and adapts to changing circumstances
+
+Cognitive Robotics often model problems based on [[Marr’s Levels of Abstraction]]
+### Approaches
+- Developmental Robotics
+- Evolutionary Robotics
+- Swarm Robotics
+- Soft Robotics
+- Neurorobotics
+### Developmental Robotics
+- interdisciplinary approach to the autonomous design of behavioral and cognitive capabilities in artificial **agents** that takes inspiration from the developmental principles and mechanisms observed in natural cognitive systems
+	- how the organization of a single robot's control system develops through experience, over time.
+	- **Developmental Psychology**
+	- ![[Pasted image 20240602124018.png]]
+	- [[Genetic Epistemology]] → Nonlinear, Stage-like development
+
+### Evolutionary Robotics
+- populations of robots that evolve over time
+- **evolutionary biology**
+- genetic algorithms
+- **Population:** A set of candidate solutions (robot designs or behaviors) that evolve over time.
+- **Selection:** Choosing the best-performing candidates based on their **fitness scores**.
+- **Crossover (Recombination):** Combining parts of two or more candidate solutions to create new offspring.
+- **Mutation:** Randomly altering parts of candidate solutions to introduce variability.
+- **Generations:** Iterations of the evolutionary process, with each generation representing a new set of candidate solutions.

content/artificial intelligence/cognitive robotics/Human-Robot Interaction (HRI).md

@@ -0,0 +1,7 @@
+Application of [[Social Robotics]] and language models to human-robot interaction scenario
+
+- Speech Recognition
+- Action Recognition and intention reading
+- Trust and acceptability
+- Emotion Recognition
+- Long-Term Interaction

content/artificial intelligence/cognitive robotics/Robotics.md

@@ -0,0 +1,15 @@
+## Robot
+- An autonomous system which exists in the physical world, can sense its environment, and act on it to achieve a goal
+	- a physical **Intelligent Agent** . 
+	- ![[Artificial Intelligence#Intelligent Agent]]
+- types
+	- humanoid
+	- android
+	- biomimetic
+	- industrial
+	- medical
+	- assistive
+
+## Robotics
+- [[Cognitive Robotics]]
+- [[Social Robotics]]

content/artificial intelligence/computer vision/Binary Segmentation.md

@@ -0,0 +1,18 @@
+### Thresholding based
+- For a greyscale image, it often suffices to apply a threshold value to the image function to segment the image
+- **b**(x,y) = (**g**(x,y) <*T*)
+	- *T* is the threshold greyscale value
+	- **g**(x,y) is the image function
+	- **b**(x,y) is a binary image
+		- image with 2 values - 1s and 0s
+		- foreground and background will be shaded with 1s and 0s respectively
+- need to choose threshold with care
+	- **automatic threshold selection** 
+		- several threshold selection algorithms
+		- one method is to use image histogram
+			- Threshold placed where rate of change of segmented area is smallest
+### Adaptive Thresholding
+- where a simple **global** threshold value is not good enough, use **Adaptive thresholding** (**local** thresholds)
+- thresholds are localized
+- an estimate of background shading can be used 
+- perform [[Image smoothing]] using something like a  [[Gaussian Smoothing]], and then subtract the resulting estimate of background variance from the original image, giving a background corrected image. Then perform thresholding over this.

content/artificial intelligence/computer vision/Computer Vision.md

@@ -0,0 +1,7 @@
+
+- [[Image Processing]]
+- [[Image Segmentation]]
+- [[Object Recognition]]
+- [[Object Tracking]]
+- [[Face Detection]] & [[Face Recognition]]
+- [[Stereo Vision]]

content/artificial intelligence/computer vision/Face Detection.md

@@ -0,0 +1,57 @@
+- Find Faces → Discriminate between faces/non faces
+- uses
+	- preprocessing step to [[Face Recognition]]
+	- Autofocus, Exposure, color balance in photography
+	- surveillance, biometrics, monitoring
+- invariant to 
+	- scale
+	- rotation
+	- illumination
+
+## Haar Features
+
+> a.k.a. Viola Jones Algorithm
+
+slide windows of different scales across the image, extract features and classify the features as face/non-face
+### Haar Filters:
+- Images are convolved with a set of Haar filters (24x24 base size) across different filter scales
+- ![[Pasted image 20240601201603.png]]
+- Haar features are sensitive to direction of the filters
+- The filters are themselves simple black and white (+1 and -1) filters
+	- the convolution can be simplified to simple summation
+	- ![[Pasted image 20240601171140.png]]
+	- filters are meant to capture certain facial features
+### Integral Image
+- table that holds the inclusive sum of all pixel intensity values to the left and top of a given pixel 
+- 2D cumulative sum
+- ![[Pasted image 20240601171558.png]]
+- integral images Allows for fast summation of image intensity of arbitrary rectangles
+- ![[Pasted image 20240601171538.png]]
+- Can be used to quickly compute Haar filters
+- ![[Pasted image 20240601171727.png]]
+- The computation cost remains constant regardless of the size of the Haar filter
+###  Classification
+- Haar features around a pixel need to be classified as face/non-face → [[Classification]] problem
+	- [[Nearest Neighbour Classifier]] 
+		- large dataset needed for robustness
+		- slow and computationally expensive
+			- large dataset comparison
+			- have to do it for every scale
+	- [[Support Vector Classifier]]
+- Large number of features
+	- use Adaboost and Cascading to optimize classification
+#### Adaboost
+
+> see [[Adaboost]]
+
+- eliminate redundant/irrelevant features
+- **weak classifier**: a relevant feature that does better than random guessing
+- Adaboost constructs a **strong classifier** as a linear combination of weak classifiers
+#### Cascading
+
+- Strong features are formed into a binary classifier
+	- +ve matches are sent along to the next feature
+	- -ve matches are rejected 
+- Reduces the amount of computation time spent on false windows
+
+![[Pasted image 20240601201056.png]]

content/artificial intelligence/computer vision/Image Segmentation.md

@@ -0,0 +1,50 @@
+- partition image to segments
+- group pixels with similar visual characteristics
+	- the specificity or scale is dependent on the domain
+- segmentation in humans
+	- likely to be different
+	- subjective
+	- ill-defined problem
+- For gray images
+	- [[Binary Segmentation]]
+	- [[Active Contours]]
+## Theory
+- [[Gestalt Psychology]]
+## Approaches
+- **Top-down**
+	- pixels belong together because they come from the same object
+	- in line with [[Gestalt Psychology]]
+- **Bottom-up**
+	- pixels belong together because they look similar
+	- Most techniques are bottom-up
+### Clustering based
+- Cluster pixels based on its visual characteristics
+- Each pixel can be seen as a feature vector of 
+	- brightness
+	- colour (R,G,B channels)
+	- position
+	- depth
+	- motion
+	- texture
+	- material
+- Map the pixels into a feature space 
+- Pixel Similarity 
+	- Dissimilarity or distance between features → [[L2 Norm]]
+- Using pixel similarity/dissimilarity, cluster the features such that similar pixels cluster together
+- each cluster is an image segment
+- disjoint regions could belong to the same cluster
+	- using position as a feature could discourage this
+- Algorithms
+	- [[K-means Clustering]] based segmentation
+	- **Model-Free** : [[Mean shift Clustering]] based segmentation
+	- **Probabilistic** : [[Gaussian Mixture Models]] based segmentation
+		- **foreground-background classification**
+			- pixels are background most of the time. So gaussians with large scale **ω** and small **σ** are background.
+			- take the ratio of **ω/σ**
+#### Graph based
+- images as graphs → G = (V,E)
+	- a vertex for each pixel, and its features, like in clustering
+	- an edge → weighted by the **affinity** or similarity between vertices
+- Algorithms
+	-  [[Graph Cut Algorithm]] based Segmentation
+		- Each subgraph is an image segment

content/artificial intelligence/computer vision/deep computer vision/Convolutional Neural Networks.md

@@ -0,0 +1,15 @@
+a.k.a **CNN** or **ConvNets**
+## Sparse Interactions
+Convolutional networks typically have sparse interactions, which is accomplished by making the kernel smaller than the input. For example, when processing an image, the input image might have thousands or millions of pixels, but we can detect small, meaningful features such as edges with kernels that occupy only tens or hundreds of pixels.
+## Parameter Sharing
+Parameter sharing refers to using the same parameter for more than one function in a model. Ina convolutional neural net, each member of the kernel is used at every position of the input. The parameter sharing used by the convolution operation means that rather than learning a separate set of parameters for every location, we learn only one set
+
+### filters, convolutions, padding, strides, pooling
+
+## CNN architectures
+[[Residual Learning]] or ResNets
+
+
+> [[Convolution]]
+
+#todo

content/artificial intelligence/computer vision/deep computer vision/Deep Computer Vision.md

@@ -0,0 +1,3 @@
+- [[Convolutional Neural Networks]]
+
+#todo