Questions tagged [architecture]

For questions related to the architecture of AI models, e.g. the architecture of neural networks.

Filter by
Sorted by
Tagged with
18
votes
2answers
425 views

Are Modular Neural Networks more effective than large, monolithic networks at any tasks?

Modular/Multiple Neural networks (MNNs) revolve around training smaller, independent networks that can feed into each other or another higher network. In principle, the hierarchical organization ...
15
votes
2answers
700 views

How can I automate the choice of the architecture of a neural network for an arbitrary problem?

Assume that I want to solve an issue with a neural network that either I can't fit to existing architectures (perceptron, Konohen, etc) or I'm simply not aware of the existence of those or I'm unable ...
10
votes
1answer
937 views

Why is the merged neural network of AlphaGo Zero more efficient than two separate neural networks?

AlphaGo Zero contains several improvements compared to its predecessors. Architectural details of Alpha Go Zero can be seen in this cheat sheet. One of those improvements is using a single neural ...
9
votes
4answers
2k views

Should neural nets be deeper the more complex the learning problem is?

I know it's not an exact science. But would you say that generally for more complicated tasks, deeper nets are required?
7
votes
2answers
90 views

Why do very deep non resnet architectures perform worse compared to shallower ones for the same iteration? Shouldn't they just train slower?

My understanding of the vanishing gradient problem in deep networks is that as backprop progresses through the layers the gradients become small, and thus training progresses slower. I'm having a hard ...
6
votes
1answer
246 views

Are there well-established ways of mixing different inputs (e.g. image and numbers)?

I am interested in the possibility of having extra input along with the main data. For instance, a medical application that would rely mostly on an image: how could one also account for sex, age, etc.?...
6
votes
0answers
96 views

How do neural network topologies affect GPU/TPU acceleration?

I was thinking about different neural network topologies for some applications. However, I am not sure how this would affect the efficiency of hardware acceleration using GPU/TPU/some other chip. If, ...
5
votes
1answer
844 views

How to create an AI to solve a word search?

This at first sounds ridiculous. Of course there is an easy way to write a program to solve a wordsearch. But what I would like to do is write a program that solves a word-search like a human. That ...
4
votes
2answers
238 views

Are there any learning algorithms as powerful as "deep" architectures?

This article suggests that deep learning is not designed to produce the universal algorithm and cannot be used to create such a complex systems. First of all it requires huge amounts of computing ...
4
votes
2answers
260 views

Are Neural Net architectures accidental discoveries?

Recently, I have been learning about new neural networks, which are used for specialized purposes, like speech recognition, image recognition, etc. The more I discover the more I get amazed by the ...
4
votes
1answer
114 views

What is a unified neural network model?

In many articles (for example, in the YOLO paper, this paper or this one), I see the term "unified" being used. I was wondering what the meaning of "unified" in this case is.
4
votes
2answers
113 views

Is a basic neural network architecture better with small datasets?

I'm currently trying to predict 1 output value with 52 input values. The problem is that I only have around 100 rows of data that I can use. Will I get more accurate results when I use a small ...
4
votes
1answer
390 views

Get the position of an object, out of an image

I have some images with a fixed background and a single object on them which is placed, in each image, at a different position on that background. I want to find a way to extract, in an unsupervised ...
4
votes
1answer
113 views

Which neural network can I use to solve this constrained optimisation problem?

Let $\mathcal{S}$ be the training data set, where each input $u^i \in \mathcal{S}$ has $d$ features. I want to design an ANN so that the cost function below is minimized (the sum of the square of ...
4
votes
1answer
170 views

What's the difference between architectures and backbones?

In the paper "ForestNet: Classifying Drivers of Deforestation in Indonesia using Deep Learning on Satellite Imagery", the authors talk about using: Feature Pyramid Networks (as the ...
3
votes
1answer
123 views

What linear rectifier is better?

What rectifier is better in general case of Convolutional Neural Network and how about empirical rules to use each type? ReLU PReLU RReLU ELU Leacky ReLU
3
votes
1answer
47 views

Are there deep neural networks that have inputs connected with deeper hidden layers?

Are there any architectures of deep neural networks that connect input neurons not only with the first hidden layer but also with deeper ones (red lines on the picture)? If so could you give some ...
3
votes
1answer
175 views

What is the input to AlphaGo's neural network?

I have been reading an article on AlphaGo and one sentence confused me a little bit, because I'm not sure what it exactly means. The article says: AlphaGo Zero only uses the black and white stones ...
3
votes
1answer
245 views

What type of reinforcement learning can I do restricted to ~200MB on an average smartphone?

This concerns a set of finite, non-trivial, combinatorial games [M] in the form of an app. A sample game can be found here. Because this is a mass market product, we can't take up too much space, ...
3
votes
1answer
2k views

What do the numbers in this CNN architecture stand for?

So I've got a neural net model (ResNet-18) and made a diagram according to the literature (https://arxiv.org/abs/1512.03385). I think I understand most of the format of the convolutional layers: ...
2
votes
1answer
50 views

How to properly use Flatten layer?

Context I'm trying to create net that will be able to recognize printed-like digits. Something like MNIST, but only for standard printing font. Images are of the size 40x40 and I'd like to put them ...
2
votes
1answer
101 views

How can we get a differentiable neural network to count things?

Imagine I have images with apples in them. I want to train a neural network which can count the number of apples in each image. BUT, I don't want to use a detector, then count the number of bounding ...
2
votes
2answers
174 views

Are artificial networks based on the perceptron design inherently limiting?

At the time when the basic building blocks of machine learning (the perceptron layer and the convolution kernel) were invented, the model of the neuron in the brain taught at the university level was ...
2
votes
2answers
158 views

In classification, how does the number of classes affect the model size and amount of data needed to train?

When solving a classification problem with neural nets, be it text or images, how does the number of classes affect the model size and amount of data needed to train? Are there any soft or hard ...
2
votes
3answers
138 views

Why does a neuron in a multi-layer network need several input connections?

For example, if I have the following architecture: Each neuron in the hidden layer has a connection from each one in the input layer. 3 x 1 Input Matrix and a 4 x 3 weight matrix (for the ...
2
votes
1answer
164 views

Neural network architecture for comparison

When someone wants to compare 2 inputs, the most widespread idea is to use a Siamese architecture. Siamese architecture is a very high level idea, and can be customized based on the problem we are ...
2
votes
1answer
61 views

Why do Transformers have a sequence limit at inference time?

As far as I understand, Transformer's time complexity increases quadratically with respect to the sequence length. As a result, during training to make training feasible, a maximum sequence limit is ...
2
votes
1answer
99 views

Tweaking a CNN for large number of input channels

I am using a CNN for function approximation using geospatial data. The input of the function I am trying to approximate consists of all the spatial distances between N location on a grid and all the ...
2
votes
0answers
27 views

Is there any way to force one input have more effect on model?

Now I am working on building a deep learning model for a regression problem. I used 50 inputs and try to add one new categorical input. The problem is that this one input is much more important than ...
2
votes
0answers
67 views

Merge two different CNN models into one

I have 2 different models with each model doing a separate function and have been trained with different weights. Is there any way I can merge these two models to get a single model. If it can be ...
2
votes
1answer
114 views

Can you explain me this CNN architecture?

I am starting to get my head around convolutional neural networks, and I have been working with the CIFAR-10 dataset and some research papers that used it. In one of these papers, they mention a ...
2
votes
2answers
55 views

Heavily mixing signal differentiation from Open Set of backgrounds via CNN

To whomever can help out, I appreciate it. I am currently attempting to detect a signal from background noise. The signal is pretty well known but the background has a lotttt of variability. I've ...
2
votes
2answers
2k views

What is the exact output of the Inception ResNet V2's feature extraction layer?

I am working with the Inception ResNet V2 model, pre-trained with ImageNet, for face recognition. However, I'm so confused about what the exact output of the feature extraction layer (i.e. the layer ...
2
votes
0answers
45 views

Multi-field text input for LSTM

I'm using LSTM to categorize medium-sized pieces of text. Each item to be categorized has several free-form text fields, in addition to several categorical fields. What is the best approach to using ...
1
vote
3answers
355 views

Are there ways to learn and practice Deep Learning without downloading and installing anything?

As per subject title, are there ways to try Deep Learning without downloading and installing anything? I'm just trying to have a feel of how this work, not really want to go through the download and ...
1
vote
2answers
542 views

How to teach a model-based reflex agent for doing some task using machine learning methods?

I would like to know how to teach an agent for performing prediction of the severity of disease and also for alerting patients using machine learning methods. I found the model-based reflex agent can ...
1
vote
1answer
57 views

CNN Architectures for local features vs global context

Kaparthy in his blog post said [this] hints at the kinds of architectures we’ll eventually explore. As an example - are very local features enough or do we need global context? I'd like to gain ...
1
vote
1answer
55 views

Where does batch normalization layers present in a neural network?

Batch normalization is a procedure widely used to train neural networks. Mean and standard deviation are calculated in this step of training. Since we train a neural network by dividing training data ...
1
vote
1answer
39 views

Are there examples of agents that use a more modest number of parameters on Pendulum (or similar environments)?

I'm looking at some baseline implementations of RL agents on the Pendulum environment. My guess was to use a relatively small neural net (~100 parameters). I'm comparing my solution with some ...
1
vote
1answer
56 views

Why are denser layers needed in computer vision neural nets?

Many neural net architectures for computer vision tasks use several convolutional layers and then several fully-connected (or dense) layers. While the reasons for using convolutional layers are clear ...
1
vote
1answer
105 views

How do the current input and the output of the previous time step get combined in an LSTM?

I am currently looking into LSTMs. I found this nice blog post, which is already very helpful, but still, there are things I don't understand, mostly because of the collapsed layers. The input $X_t$,...
1
vote
0answers
30 views

How to learn transition type in a 1-hour extended DJ Mix?

How would you design a model which learns the transitions in a given 1-hour DJ Mix? To be specific, the model should be able to learn transitions, specify the occurring time and the type (Crossfade, ...
1
vote
0answers
30 views

Under what circumstances is a fully connected layer similar to PCA?

I am reading this paper on image retrieval where the goal is to train a network that produces highly discriminative descriptors (aka embeddings) for input images. If you are familiar with facial ...
1
vote
0answers
45 views

Are there regularisation methods related only to architecture of the CNNs?

Are there any methods of regularisation of deep neural networks, particularly CNNs (or generally ANN but that will also work on CNNs) that are related only to the network's architecture and not the ...
1
vote
0answers
60 views

What exactly are deep learning primitives?

I came across the concept of "deep learning primitives" from the Nvidia talk Jetson AGX Xavier New Era Autonomous Machines (on slide 44). There doesn't seem to be a lot of articles in the ...
1
vote
0answers
42 views

How can one be sure that a particular neural network architecture would work?

Traditionally, when working with tabular data, one can be sure(or at least know) that a model works because the included features could explain a target variable, say "Price of a ticket" ...
1
vote
2answers
109 views

Should batch-normalization/dropout/activation-function layers be used after the last fully connected layer?

I am using the following architechture: 3*(fully connected -> batch normalization -> relu -> dropout) -> fully connected Should I add the ...
1
vote
0answers
45 views

Using U-NET for image semantic segmentation

I'm getting literally crazy trying to understand how U-NET works. Maybe it is very easy, but I'm stuck (and I have a terrible headache). So, I need your help. I'm going to segment MRI to find white ...
1
vote
0answers
15 views

Can Grad CAM feature maps be used for Training?

I am trying to recreate the architecture of the following paper: https://arxiv.org/pdf/1807.03058.pdf Can someone help me in explaining how are the feature maps coming out of the output of GradCam ...
1
vote
0answers
32 views

Neural Network training on one example to try overfitting leads to strange predictions

tldr; if I train the network on 1 training example, the outcome sometimes makes no sense at all, sometimes is as expected. If I train it on more examples and higher iterations, the network, which ...