Skip to main content

All Questions

Filter by
Sorted by
Tagged with
1 vote
0 answers
22 views

KL annealing for a VAE does not work, what now?

I am trying to train a variational auto-encoder where x ≈ f_VAE(x) = x_hat. In my real problem, I have 100-400 dimensional data that I would like to compress to ...
Patrickens's user avatar
-1 votes
0 answers
12 views

Relation between word embeddings to a more general encoding (such as in VAE, contrastive learning, and intermediate layers)

Contrastive learning finds a representation in an m-dimensional space, where similar examples will be close to each other and dissimilar examples will lie far away. Similarly, word embeddings are a ...
Liubove's user avatar
  • 11
1 vote
1 answer
35 views

Does anyone use Statistical Energy to monitor generative AI training?

Statistical Energy (Szekely & Rizzo, 2013 or Aslan & Zech, 2005) can be used as a statistical test of whether two distributions are the same or different. It works particularly well on high ...
tkw954's user avatar
  • 111
0 votes
0 answers
15 views

Which generative model architecture (and loss function) should I use to train on a dataset of 3D arrays of embedded tokens?

I've curated a dataset of player-made Minecraft builds. Each unique Minecraft block is tokenized and treated as a unique "word" like in NLP. I've trained a Skip-Gram model on the dataset (...
schmixi's user avatar
0 votes
0 answers
22 views

VAE to compare two datasets

I have two cell-by-gene matrices, each representing gene counts for cells in sample 1 and sample 2. I'm interested in identifying common gene expression patterns across both samples. These patterns ...
Yulia Kentieva's user avatar
2 votes
1 answer
69 views

Question regarding the ELBO decomposition proposed by Hoffman&Johnson

recently I'm trying to read a paper by Hoffman and Johnson discussing an alternative decomposition of ELBO in variational autoencoders. In formula (9) and (10) of their original paper, they proposed ...
Izzy Tse's user avatar
1 vote
1 answer
47 views

MLP Gaussian Decoder in VAE

My question concerns the paper arxiv.org/pdf/1312.6114. I want to know why they proposed to use MLP Gaussian decoder with parameters given by the MLP transformation of the z variable as the likelihood ...
piero's user avatar
  • 133
2 votes
1 answer
179 views

Deriving ELBO for Diffusion Models

I am trying to read through the proof of ELBO for diffusion models on pg. 8 of this paper. However, I do not see how the author arrived at Eqn (45) from Eqn (44). Specifically, I do not know how they ...
Nikhil Sridhar's user avatar
1 vote
0 answers
29 views

Any tutorials/courses to learn variational autoencoders on tabular data?

I aim to use variational autoencoders (VAE) to find interpretable latent spaces for genetic data. So, I need to understand how they work, what activation function to use, etc. But all tutorials and ...
Yulia Kentieva's user avatar
2 votes
0 answers
83 views

VAE suffers from posterior collapse under all hyper parameters [closed]

I am trying to find a low-dimensional latent space representation for a bunch of simulated data. No matter what VAE architecture I try and no matter how I tweak it, the output of the VAE is always the ...
Patrickens's user avatar
2 votes
1 answer
93 views

Derivation of the consistency term in the DDPM Evidence Lower Bound (ELBO) [closed]

I have been studying diffusion models from this tutorial: https://arxiv.org/abs/2403.18103 and trying to derive all results as I read it. Although this tutorial is very comprehensive, it skips many of ...
ahxmeds's user avatar
  • 31
0 votes
0 answers
104 views

How to solve the exploding gradient problem in VAE training?

I was trying to implement VAE on the CelebA dataset inspired by the Tensorflow implementation of MNIST. I have tried varying batch size but there seems to be no effect from that. The image formed is ...
Vedant Bhardwaj's user avatar
0 votes
0 answers
25 views

In VAE, why not maximize D[q(z)||p(z|x)] instead?

In VAE, the goal is to maximize $\log p(x)$, where $x$ is the data and $p(\cdot)$ is a parameterized distribution. For any distribution $q(z)$, the following identity holds, $$ \log p(x)=D[q(z)||p(z|x)...
John Ao's user avatar
0 votes
1 answer
112 views

Why do we add noise when training VAEs, instead of just insisting each training batch is Normally distributed at hidden layer?

I had come to think of VAEs as having 2 loss terms - A reconstruction loss that tries to ensure that the input matches the output (just as a regular Auto Encoder), and a KL-Divergence Loss term that ...
Steven's user avatar
  • 1
1 vote
1 answer
218 views

Trying to understand some derivation in the paper: Deep Unsupervised Learning using Nonequilibrium Thermodynamics

I have recently been learning about diffusion models and trying to derive all the results in the paper by Sohl-Dickstein, et. al, "Deep Unsupervised Learning using Nonequilibrium Thermodynamics&...
ahxmeds's user avatar
  • 31
0 votes
1 answer
79 views

How to remove random noise from an image (denoising)?

When adding noise to an image, for instance, is the noise added evenly random (equally likely values within some range), or random but following some distribution (like the normal distribution)? Then,...
James's user avatar
  • 157
3 votes
1 answer
360 views

How to perform latent space Interpolation between two images?

I have a variational convolutional autoencoder that has trained on 2 images and outputs a linear interpolation (inserted at the bottleneck stage) between those 2 input images. However, the result ...
James's user avatar
  • 157
3 votes
0 answers
66 views

How random should an untrained generative AI output really be?

I am developing a particular implementation of VAE, and, how usually one does while implementing any architecture, I passed a random input to the model to test if everything worked fine (e.g. check ...
GPU'njoyer's user avatar
4 votes
3 answers
221 views

In the VAE, why is $z \sim \mathcal{N}(\mu, \sigma^2)$ equivalent to $z = \mu + \sigma \odot \epsilon$?

In the reparameterization trick of a Variational Autoencoder (VAE), instead of sampling noise $z$ from $z \sim \mathcal{N}(\mu, \sigma^2)$, we can use a different method: $z = \mu + \sigma \odot \...
user avatar
0 votes
0 answers
26 views

Is there a way to mix the Importance weighted VAE with the Beta-VAE?

Is there a way to mix the Importance weighted VAE with the Beta-VAE? It seems that we cannot separate out the KL-term from the IWAE-ELBO, hence we cannot multiply it with Beta. On the other hand, it ...
Clara's user avatar
  • 11
2 votes
2 answers
288 views

What is the meaning of log p(x) in VAE math and why is it constant

I was reading the article on medium, where the author cites this equation for Variational Inference: \begin{align*} \text{KL}(q(z|x^{(i)})||p(z|x^{(i)})) &= \int_z q(z|x^{(i)})\text{log}\frac{q(z|...
Kiran Manicka's user avatar
1 vote
1 answer
198 views

How does using the ELBO in VAEs make the problem tractable?

I'm studying Variational Autoencoders and a lot of the literature says that the posterior is intractable because the marginal distribution p(x) is intractable since the space of z is so large we ...
Kiran Manicka's user avatar
0 votes
1 answer
125 views

Are there cases where Variational Auto-Encoders (VAE's) are preferred to other techniques?

The best reason I have seen for using variational autoencoders is when dealing with sparse data. The Gaussian noise "splats" out the input distribution (see this StackExchange answer). ...
programjames's user avatar
0 votes
0 answers
60 views

What best practices for VAE do you know?

The data is binary voxel data of shape (60, 36, 60). I want to compress such data into ...
Renat Abdrakhmanov's user avatar
0 votes
1 answer
832 views

Why does the latent space in Stable Diffusion have a shape of 64x64x3?

Since the encoding is performed by a Variational Autoencoder, the VAE encoder must output some mean and log variance that we can ...
Renat Abdrakhmanov's user avatar
0 votes
0 answers
24 views

How to apply Latent Diffusion for 3D Binary Voxel Data?

Suppose we have a voxel of shape (60, 36, 60) with values 0 or 1 (1-occupied, 0-empty). What is the possible architecture of latent diffusion?
Renat Abdrakhmanov's user avatar
0 votes
0 answers
39 views

Which main steps should I consider in order to successfully use a VAE for Anomaly Detection?

I am thinking about using the variational autoencoder model for anomaly detection . I have an Android Logs dataset. As the logs generated are a representative of time series type of data I thought ...
MLenthusiast's user avatar
0 votes
0 answers
83 views

VAE ( variational autoencoder) for timeseries anomaly detection ,

I am implementing VAE based anomaly detection for multivariate timeseries using keras, I have ELBO (Evidence lower bound) which is combination of $$-\ D_{KL}\left({\ q}_\varphi\left(z\middle| x^i\...
mmv_87's user avatar
  • 1
1 vote
1 answer
99 views

Variational Lower Bound in VAE for Gaussian latent prior

From Bishop's recent book on Deep Learning, it says the ELBO for Gaussian latent prior can be approximated by $\frac{1}{L}\sum_{l=1}^L \ln p(x_n|z_n^l,w) + KL(q(z_n|x_n,\phi)||p(z_n))$ where $n$ are ...
piero's user avatar
  • 133
1 vote
1 answer
190 views

What is an information bottleneck in the context of ELBO and Hierarchical VAEs?

These slides (slide number 26) mention that the ELBO enforces an information bottleneck at the latent variables z which make it prone to bad local minima. Can you please explain what they mean by that?...
ketan dhanuka's user avatar
1 vote
3 answers
191 views

What is the difference between q and p in Statistical Notation(used in VAE)?

I'm looking at general visuals of Variational Autoencoders and I'm seeing that the encoder is typically expressed as q(z|x) with phi as a subscript while the decoder is p(x|z) with theta as a ...
Kiran Manicka's user avatar
1 vote
1 answer
91 views

Is there a way to reward my Variational-Auto encoder for using less colors while still letting it make creative decisions?

So recently I have been trying to program a Tensorflow and Keras based model that can animate pixel art characters. I use a Variation Auto Encoder with Convolutions, Dense Layers and Upsampling 2d ...
Max's user avatar
  • 13
1 vote
0 answers
93 views

Pointers to (deep) latent variable models that admit analytical approximations

I am aware that there is a plethora of deep generative models out there (e.g. variational autoencoders (VAE), GANs) that can model high-dimensional data as the images of latent variables under a non-...
ngiann's user avatar
  • 111
0 votes
0 answers
41 views

VAE for Motion Sequence Generation - Convergence Issue with Scheduled Sampling

I have implemented a Variational Autoencoder (VAE) in PyTorch for motion sequence generation using human pose data (joint angles and angular velocities in radians) from the CMU dataset. The VAE ...
RTn's user avatar
  • 1
1 vote
0 answers
29 views

Toy dataset: Radial VAE

I'm evaluating disentanglement in toy datasets seeing as we have such little understanding of the phenomena. I'm using various tools from differential geometry. Now I want to train a VAE on the ...
John Miller's user avatar
1 vote
1 answer
571 views

Variational Autoencoder (VAE) Multiclass Interpolation

I'm working on a variational autoencoder (VAE) with 20 different classes in my training data. I've successfully trained the VAE and can sample from the latent space to generate data points. However, I ...
charactercapital's user avatar
2 votes
1 answer
988 views

What exactly is meant by variational distribution?

What specifically does the term "variational distribution" refer to? The encoder of Variational autoencoder? Forward process of denoising diffusion probabilistic models? Images or latent ...
diffusion stable's user avatar
0 votes
1 answer
126 views

Confusion over taking gradients in Variational Autoencoders (VAE)

I am confused as to when to hold certain parameters constant in a VAE. I will explain with a concrete example. We can write $\operatorname{ELBO}(\phi, \theta) = \mathbb{E}_{q_{\phi}(z)}\left[\log \...
Decaying Tails's user avatar
2 votes
1 answer
143 views

Why can Variational Autoencoders (VAEs) approximate arbitrary distributions?

I am trying to reason to myself why is it that VAEs can approximate arbitrary probability distributions even though 𝑞𝜙(𝑧|𝑥) and 𝑝𝜃(𝑥|𝑧) are Gaussian. I understand that the parameters are ...
Decaying Tails's user avatar
0 votes
2 answers
163 views

How to expand reconstruction error to mean squared error in Variational AutoEncoder? [closed]

How to expand reconstruction error to mean squared error when it is $\mathbb{E}_{z\sim q_{\phi}(z|x)}[\log p_\theta(x|z)]$? [reconstruction error] $\mathbb{E}_{z\sim q_{\phi}(z|x)}[\log p_\theta(x|z)]$...
diffusion stable's user avatar
1 vote
1 answer
192 views

Comparison of the two alternative forms for the KL divergence [closed]

On page 468 of 'Pattern Recognition and Machine Learning', what does 'the same variables given by the product of two independent univariate Gaussian distributions' mean? The PDF says, The green ...
diffusion stable's user avatar
3 votes
0 answers
178 views

Why is $p(x)=\int p(x,z) dz$ intractable for continuous $z$ in VAE?

In VAE we use the importance sampling trick to use $q_\phi(z|x)$ to help maximize $\log p_\theta(x)=\log \int p_\theta(x,z)dz\ge \int q_\phi(z|x)\log \frac{p_\theta(x,z)}{q_\phi(z|x)} dz$. Meanwhile, ...
hjenryin's user avatar
1 vote
2 answers
65 views

How to apply backpropagation when one layer of the network is a call-only function (no gradient)?

I worked with Feed Forward Neural Network and VAE and understood backpropagation algorithm. Now I build a VAE network, one layer of it is a very complex vector-to-vector function $f(x)$ (a general '...
whitegreen's user avatar
0 votes
1 answer
234 views

Any suggestion on what I should try to get this cVAE model working for random generation of new molecules?

I am using a cVAE model to generate new structures of molecules. After successfully training the VAE model I am able to get proper reconstructions of the training set. While I am also able to generate ...
Formal_this's user avatar
0 votes
1 answer
33 views

How to convert my test data in the same dimensionality as my train data

I have trained a VAE with jpg images. My latent space dimension has 768 features and when plotting the latent space it looks like this: However, when I use the scikit learn tool LDA (Linear ...
Dude Rar's user avatar
1 vote
2 answers
505 views

Why optimise log p(x) rather than log p(x|z) in a Variational AutoEncoder?

Background The loss function in a Variational AutoEncoder is the Evidence Lower Bound (ELBO): $\mathbb{E}_q[log$ $p(x|z)] - KL[q(z)||p(z)]$ And has this inequality: $log$ $p(x) \ge \mathbb{E}_q[log$ $...
Titus Buckworth's user avatar
0 votes
1 answer
59 views

Why is the variational lower bound is easier to compute than the original marginal distribution?

Why is the ELBO of $p(x)=\int p(x|z)p(z)\mathrm{d}z$ easier to compute/estimate than the expression itself? Can we compute this quantity itself through sampling in the same way? I understanding that ...
Hanhan Li's user avatar
  • 101
0 votes
0 answers
230 views

Using a VQ-VAE encoder's output to condition another model

I've trained a VQ-VAE on images that I want to use to condition another model (latent diffusion). The shape of my encoder's output is (4, 256, 56, 56) and I'm getting it using: ...
jbm's user avatar
  • 101
-1 votes
1 answer
243 views

Combination of VAE with GAN

I am going to implement a lecture which it aims to generate new images. It uses a variational autoencoder to produce latent vector and then feed it to a gan network as input. My question is, in ...
Pedram Yazdipoor's user avatar
0 votes
1 answer
515 views

How to generate new data using VAE?

I have built the following function which takes as input some data and runs a VAE on them: ...
quant's user avatar
  • 101