Receptive Fields - Explained

CodeEmporium · Advanced ·🔢 Mathematical Foundations ·1y ago

Key Takeaways

The video explains receptive fields in the human brain, specifically center surround receptive fields, and their relation to computer vision and edge detection. It delves into the physiology of retinal ganglion cells and their connections to the center and surround regions, discussing on-center and off-center receptive fields.

Full Transcript

Greetings fellow learners. In this video we are going to talk about receptive fields. So let's start with a broad definition. A receptive field is a region of space that affects the cell being studied. So now if we were to take some specific example of this. Let's say that we have this mammal that is looking at some point in a visual scene and this back layer over here of the eye is known as the retina. So let's say that in this retina it composes of many types of cells. One such cell is the retinal ganglen cell that'll connect to the brain. And let's say that this is the cell that's being studied. Now this specific retinal ganglen cell is going to be affected by this ray of light which is coming from this source over here in the visual scene. And so what we can say is that for this retinal ganglen cell that's over here the receptive field is this point that is this dot here in the visual scene. Now this is one way to look at what a receptive field is. Now in another case we can say that the receptive field could be neurons too. So for example if we take let's say this part of the retina and we blow it up we might get something that looks like this. So let's say that this here is going to be the front of the retina. This is the back of the retina. These are photo receptors. This is a bipolar cell. These are horizontal cells. And the bipolar cell is going to be connected to the cell that we want to study, which is the retinal ganglen cell. Now, we know that for a receptive field, the receptive field is going to be that region that affects this cell in question, which is the retinal ganglion cell. And so, that's effectively going to be this bipolar cell, this horizontal cell, and the blue photo receptors as well as the pink or red photo receptors. All of this will be constituted in the receptive fields of the retinal ganglen cell. And interestingly in the retina, this is actually how the cells are organized. The receptive field is going to be in two concentric circles, especially particularly at this photo receptor layer. So the inner concentric circle is known as the center and the outer concentric circle is known as the surround. And how do we get an inner circle and an outer circle? Well, it depends on how they are physiologically that is actually the neurons are actually going to be synapsing into the retinal ganglen cell. So the center cells over here they are going to directly synapse onto the bipolar cell which will directly synapse onto the ganglen cell. The surround cells are going to synapse into horizontal cells which will then synapse into bipolar cells which will then synapse into the retinal ganglion cell that is being studied. So one is the center is directly connected whereas the receptive field surround is indirectly connected via a horizontal cell. That's the main difference. Now with this physiology a retinal ganglen cell can have two types of receptive fields. In one type of receptive field if light passes to the photo receptors in the center region then the ganglen cell is going to be excited that is it'll start firing very rapidly. On the other hand, if light is shown and absorbed only by the photo receptors in the surround region, then the retinal ganglen cell might be inhibited. That is the frequency of its response slows down. If this is the case, then the retinal ganglen cell is going to be considered to have an on center off surround receptive field. Conversely, a retinal ganglen cell could also have another type of receptive field where if light falls on the photo receptors in the center region, they are going to inhibit the retinal ganglen cell. And if right light falls on the surround region, they could excite the retinal ganglen cell. If the retinal gangleen cell behaves in this way, then it is set to have an off center on surround receptive field. Now, if you're wondering why is it only these two cases and why can't there be a case where you know light falls both on the center or surround and they both excite or they both inhibit the retinal ganglen cell. It's mostly due to the functioning of the horizontal cell in question where it will essentially ensure that the effect of the center and surround is opposite to each other. And you can look up this for more information on the concept of lateral inhibition. In order to crystallize the understanding here, I'm going to share a demo here on YouTube. So first we're going to take a look at an on center ganglen cell receptive field. In this case, this is the two concentric receptive fields. This here represents the frequency of action potentials. Action potentials is like the firing of a neuron to communicate information and this is going to be each of these bars vertical bars corresponds to a firing. So the rate of firing is going to be this red bar which can go up or down if the frequency of firing of the action potentials changes. So let's now play this video and we'll see that whenever we shine light on the center region of an on center ganglen cell you'll see that this frequency increases quite rapidly. On the other hand if it's only shown on this region the surround region the frequency is going to decrease. So this shows some amount of like information. Now if we shine light only entirely on the center region of an onenter ganglen cell the frequency of action potentials becomes maximum and so that means that the retinal ganglen cell is maximally excited. On the other hand if we show light only on the surround regions over here the retinal ganglen cell is maximally inhibited which means that it doesn't fire at all. And interestingly, if you shine light on both the center and the surround, the ganglen cell is not really going to transmit any more information than it usually would as if there was no light there itself. So this shows that you know there if there's a contrast of light falling on the center and surround more information is sent from the ganglen cell and eventually to the brain for visual processing. Now, if we look at an off center ganglen cell, we're going to see the exact opposite of this case, right? So, if we shine light at the center of an off-center ganglen cell, you'll see that the the ganglen cell is going to be inhibited and that means that this frequency is going to go down. On the other hand, if you shine it only on this surround region, then that means that the frequency is going to go much higher as the ganglen cell is excited. And here you can see that if we shine light only on the center of an off-center ganglen cell, the response of the ganglen cell is maximally inhibited. And if you do the converse of you know light is only going to be on the surround this makes the retinal ganglen cell maximally excited starts firing rapidly and sends this information to the brain and similar to the on center case too if we shine light entirely on the ganglen cell well it's not really going to send any different of a response or much different of a response compared to no light at all. Now why does this actually even important is that well we'll see here that the retinal ganglen cell actually helps in edge detection at least an early form of visual processing of edges. So if this is an edge and we can see that when we drag this ganglen cell down in towards the edge there as long as the edge falls on the receptive field at all whether it's the center or surround there will be a change in the firing rate of the ganglen cell. This conveys information that is going to be passed on to the brain for further processing. And so ganglen cell can be used to detect edges. And I think this here is going to be a nice graph that shows exactly that where we'll see the most activity though when the center is completely in the light and just completely in the right and in the it's going to be the least active or you know it's going to be most inhibited in this case when the center is just entirely outside the light. At least this is for on center ganglen cells and overall they can detect edges. Quiz time. Have you been paying attention? Let's quiz you to find out. Why do photo receptors in the center region of a retinal ganglen cell's receptive field behave oppositely to those in the surround? A. Horizontal cells invert the surround signal. B. surround photo receptors use a reverse acting pigment. C amocrine cells flip the center signal or D more photo receptors converge in the center. I'll give you a few seconds to answer this question. The correct answer is A. But can you tell me why? Give your reasoning down below and let's have a discussion. And at this point, if you think I do deserve it, please do consider giving this video a like because it will help me out a lot. And that's going to do it for quiz time and this video. But before we go, let's generate a summary. So in this video, we started with the general definition of a receptive field. It's a region of space that affects the cell that's being studied. And this receptive field could be something like a point on the visual scene for a specific like retinal ganglen cell here. Or we can also define a receptive field in terms of neurons and cells. In this case, all of these photo receptors constitute the receptive fields of the retinal ganglen cell in question. And specifically for the retinal ganglen cell neurons here, they are organized in concentric circles of a center region and a surround region depending on if they're directly or indirectly connected to a bipolar cell. Now retinal ganglen cells can have two types of respective fields depending on how they react to light falling on their photo receptors. So if they are activated by light falling on the center and inhibited by light falling on the surround, they are known as on center off surround receptive fields. Conversely, if they are inhibited by light falling on the center and excited by light falling in the surround, then the retinal ganglen cell has an offcenter on surround receptive field. We also took a look at some demos for both of these types of receptive fields and how they interact with light either in just the center only the surround completely in the center completely in the surround or both cases. And we also noted the frequency of action potentials over here at a slightly more higher level. We also saw how these ganglen cells could be used to detect edges as their response and activity mainly changes when an edge comes across the receptive field entirely of this ganglen cell. And this edge detection can be used further in the brain for further visual processing. Now that's all that we have for today. Thank you all so much for watching. If you like this video, please do consider giving this video a like, share, and I will see you in the next one.

Original Description

We talk about center surround receptive fields in the human brain. Linked to computer vision. ABOUT ME ⭕ Subscribe: https://www.youtube.com/c/CodeEmporium?sub_confirmation=1 📚 Medium Blog: https://medium.com/@dataemporium 💻 Github: https://github.com/ajhalthor 👔 LinkedIn: https://www.linkedin.com/in/ajay-halthor-477974bb/ RESOURCES [1 📚] Slides: https://link.excalidraw.com/p/readonly/9J2Eflz4aM5rNgjPOz8X [2 📚] Main paper in 1953 that discovered these center surround receptive fields: https://journals.physiology.org/doi/abs/10.1152/jn.1953.16.1.37 [3 📚] Video demo presented in video: https://www.youtube.com/watch?v=ZR7LzRAXNSw&list=LL&index=2 [4 📚] Summary notes on receptive fields: https://www.memcode.com/courses/1965 PLAYLISTS FROM MY CHANNEL ⭕ Reinforcement Learning: https://youtube.com/playlist?list=PLTl9hO2Oobd9kS--NgVz0EPNyEmygV1Ha&si=AuThDZJwG19cgTA8 Natural Language Processing: https://youtube.com/playlist?list=PLTl9hO2Oobd_bzXUpzKMKA3liq2kj6LfE&si=LsVy8RDPu8jeO-cc ⭕ Transformers from Scratch: https://youtube.com/playlist?list=PLTl9hO2Oobd_bzXUpzKMKA3liq2kj6LfE ⭕ ChatGPT Playlist: https://youtube.com/playlist?list=PLTl9hO2Oobd9coYT6XsTraTBo4pL1j4HJ ⭕ Convolutional Neural Networks: https://youtube.com/playlist?list=PLTl9hO2Oobd9U0XHz62Lw6EgIMkQpfz74 ⭕ The Math You Should Know : https://youtube.com/playlist?list=PLTl9hO2Oobd-_5sGLnbgE8Poer1Xjzz4h ⭕ Probability Theory for Machine Learning: https://youtube.com/playlist?list=PLTl9hO2Oobd9bPcq0fj91Jgk_-h1H_W3V ⭕ Coding Machine Learning: https://youtube.com/playlist?list=PLTl9hO2Oobd82vcsOnvCNzxrZOlrz3RiD MATH COURSES (7 day free trial) 📕 Mathematics for Machine Learning: https://imp.i384100.net/MathML 📕 Calculus: https://imp.i384100.net/Calculus 📕 Statistics for Data Science: https://imp.i384100.net/AdvancedStatistics 📕 Bayesian Statistics: https://imp.i384100.net/BayesianStatistics 📕 Linear Algebra: https://imp.i384100.net/LinearAlgebra 📕 Probability: https://imp.i384100.net/Probability OTHER
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from CodeEmporium · CodeEmporium · 0 of 60

← Previous Next →
1 Linear Regression and Multiple Regression
Linear Regression and Multiple Regression
CodeEmporium
2 Logistic Regression - THE MATH YOU SHOULD KNOW!
Logistic Regression - THE MATH YOU SHOULD KNOW!
CodeEmporium
3 Generative Adversarial Networks - FUTURISTIC & FUN AI !
Generative Adversarial Networks - FUTURISTIC & FUN AI !
CodeEmporium
4 Deep Learning on the Cloud - GPU TO LEARN FASTER
Deep Learning on the Cloud - GPU TO LEARN FASTER
CodeEmporium
5 Deep Mind's AlphaGo Zero - EXPLAINED
Deep Mind's AlphaGo Zero - EXPLAINED
CodeEmporium
6 Mask Region based Convolution Neural Networks - EXPLAINED!
Mask Region based Convolution Neural Networks - EXPLAINED!
CodeEmporium
7 Attention in Neural Networks
Attention in Neural Networks
CodeEmporium
8 Depthwise Separable Convolution - A FASTER CONVOLUTION!
Depthwise Separable Convolution - A FASTER CONVOLUTION!
CodeEmporium
9 One Neural network learns EVERYTHING ?!
One Neural network learns EVERYTHING ?!
CodeEmporium
10 Neural Voice Cloning
Neural Voice Cloning
CodeEmporium
11 AI creates Image Classifiers…by DRAWING?
AI creates Image Classifiers…by DRAWING?
CodeEmporium
12 Unpaired Image-Image Translation using CycleGANs
Unpaired Image-Image Translation using CycleGANs
CodeEmporium
13 K-Means Clustering - EXPLAINED!
K-Means Clustering - EXPLAINED!
CodeEmporium
14 Random Forest Classification
Random Forest Classification
CodeEmporium
15 Data Science in Finance
Data Science in Finance
CodeEmporium
16 Hypothesis testing with Applications in Data Science
Hypothesis testing with Applications in Data Science
CodeEmporium
17 A/B Testing - Simply Explained
A/B Testing - Simply Explained
CodeEmporium
18 The Kernel Trick - THE MATH YOU SHOULD KNOW!
The Kernel Trick - THE MATH YOU SHOULD KNOW!
CodeEmporium
19 Support Vector Machines - THE MATH YOU  SHOULD KNOW
Support Vector Machines - THE MATH YOU SHOULD KNOW
CodeEmporium
20 Principal Component Analysis (PCA) - THE MATH YOU SHOULD KNOW!
Principal Component Analysis (PCA) - THE MATH YOU SHOULD KNOW!
CodeEmporium
21 History of Calculus - Animated
History of Calculus - Animated
CodeEmporium
22 Curiosity in AI
Curiosity in AI
CodeEmporium
23 DropBlock - A BETTER DROPOUT for Neural Networks
DropBlock - A BETTER DROPOUT for Neural Networks
CodeEmporium
24 Autoencoders - EXPLAINED
Autoencoders - EXPLAINED
CodeEmporium
25 Recurrent Neural Networks - EXPLAINED!
Recurrent Neural Networks - EXPLAINED!
CodeEmporium
26 LSTM Networks - EXPLAINED!
LSTM Networks - EXPLAINED!
CodeEmporium
27 Building an Image Captioner with Neural Networks
Building an Image Captioner with Neural Networks
CodeEmporium
28 10 Machine Learning Questions - ANSWERED!
10 Machine Learning Questions - ANSWERED!
CodeEmporium
29 How do neural networks work?
How do neural networks work?
CodeEmporium
30 Evolution of Face Generation |  Evolution of GANs
Evolution of Face Generation | Evolution of GANs
CodeEmporium
31 How does Google Translate's AI work?
How does Google Translate's AI work?
CodeEmporium
32 How to keep up with AI research?
How to keep up with AI research?
CodeEmporium
33 How does YouTube recommend videos? - AI EXPLAINED!
How does YouTube recommend videos? - AI EXPLAINED!
CodeEmporium
34 Variational Autoencoders - EXPLAINED!
Variational Autoencoders - EXPLAINED!
CodeEmporium
35 Logistic Regression - VISUALIZED!
Logistic Regression - VISUALIZED!
CodeEmporium
36 Gradient Descent - THE MATH YOU SHOULD KNOW
Gradient Descent - THE MATH YOU SHOULD KNOW
CodeEmporium
37 Boosting - EXPLAINED!
Boosting - EXPLAINED!
CodeEmporium
38 Transformer Neural Networks - EXPLAINED! (Attention is all you need)
Transformer Neural Networks - EXPLAINED! (Attention is all you need)
CodeEmporium
39 Loss Functions - EXPLAINED!
Loss Functions - EXPLAINED!
CodeEmporium
40 Optimizers - EXPLAINED!
Optimizers - EXPLAINED!
CodeEmporium
41 NLP with Neural Networks & Transformers
NLP with Neural Networks & Transformers
CodeEmporium
42 Batch Normalization - EXPLAINED!
Batch Normalization - EXPLAINED!
CodeEmporium
43 Activation Functions - EXPLAINED!
Activation Functions - EXPLAINED!
CodeEmporium
44 Data Scientist Answers Interview Questions
Data Scientist Answers Interview Questions
CodeEmporium
45 Why use GPU with Neural Networks?
Why use GPU with Neural Networks?
CodeEmporium
46 How do GPUs speed up Neural Network training?
How do GPUs speed up Neural Network training?
CodeEmporium
47 BERT Neural Network - EXPLAINED!
BERT Neural Network - EXPLAINED!
CodeEmporium
48 ConvNets Scaled Efficiently
ConvNets Scaled Efficiently
CodeEmporium
49 Transformer Neural Net makes music! (JukeboxAI)
Transformer Neural Net makes music! (JukeboxAI)
CodeEmporium
50 What do filters of Convolution Neural Network learn?
What do filters of Convolution Neural Network learn?
CodeEmporium
51 We're hosting a Machine Learning Conference!
We're hosting a Machine Learning Conference!
CodeEmporium
52 MLconfEU 2020: Machine Learning Conference for Software Engineers
MLconfEU 2020: Machine Learning Conference for Software Engineers
CodeEmporium
53 Are Neural Networks Intelligent?
Are Neural Networks Intelligent?
CodeEmporium
54 Time Series Forecasting with Machine Learning
Time Series Forecasting with Machine Learning
CodeEmporium
55 Few Shot Learning - EXPLAINED!
Few Shot Learning - EXPLAINED!
CodeEmporium
56 How does a Data Scientist Fight FRAUD?
How does a Data Scientist Fight FRAUD?
CodeEmporium
57 How would a Data Scientist analyze Customer Churn?
How would a Data Scientist analyze Customer Churn?
CodeEmporium
58 Expectations with Machine Learning
Expectations with Machine Learning
CodeEmporium
59 Why Logistic Regression DOESN'T return probabilities?!
Why Logistic Regression DOESN'T return probabilities?!
CodeEmporium
60 How you SHOULD code Machine Learning
How you SHOULD code Machine Learning
CodeEmporium

This video teaches about receptive fields, specifically center surround receptive fields, and their importance in visual processing and edge detection. It explains how retinal ganglion cells respond to light and how this relates to computer vision. By understanding receptive fields, viewers can better appreciate how the human brain processes visual information and apply this knowledge to computer vision tasks.

Key Takeaways
  1. Take a specific example of a mammal looking at a point in a visual scene
  2. Identify the retinal ganglion cell and its connections to the brain
  3. Describe the receptive field of the retinal ganglion cell as a region of space that affects the cell being studied
  4. Explain the physiology of the retinal ganglion cell and its connections to the center and surround regions
  5. Describe the two types of receptive fields: on center off surround and off center on surround
  6. Analyze how retinal ganglion cells help in edge detection by conveying information about changes in firing rate when light falls on the receptive field
💡 The receptive field of a retinal ganglion cell can be affected by photo receptors in the center and surround regions, and this has implications for computer vision and edge detection.

Related AI Lessons

Up next
How to Open OSM Files (OpenStreetMap Data)
File Extension Geeks
Watch →