Questions tagged [sequence-modeling]
Questions about the analysis of sequential data, often used to analyse audio information or to predict time series.
69 questions
0
votes
0
answers
25
views
Sequence to Sequence vs Sequence to Token
I came here to ask for some clarification about the subject that is in the topic.
Denote: Seq2Seq, Seq2Tok
What I am trying to understand if there is any use of the output sequence in Seq2Seq models ...
- 101
0
votes
0
answers
13
views
Is there another way to calculate receptive field size for temporal convolutional networks?
In the paper https://arxiv.org/pdf/1905.12853 on page 4 a TCN-model is used where the receptive field is 253 frames, kernel size is 3 and 6 residual blocks are used. As far as I know the formula for ...
1
vote
2
answers
196
views
Wouldn't residual connections in RNNs solve the vanishing/exploding gradient problem?
I was recently brushing up on my deep-learning basics and came back to RNNs. LSTMs/GRUs and the Transformer architecture were invented to solve RNN's vanishing/exploding gradient problem. I was at ...
0
votes
0
answers
25
views
Executing Multiple ML Models simultaneously on multiple cores to reduce the model building time
I have a time series forecasting problem which consist of date, item no and quantity columns. I have defined a function which takes input as data frame and forecasting period (Daily,Weekly,Monthly,...
- 1
0
votes
0
answers
65
views
Does the accuracy of a regression learner depend on the way we feed data?
Consider a plot of points as such:
As one notices, this looks like an alternating sequence. Further, it can be divided into two subsequences as $a_{\text{odd}}$ and $a_{\text{even}}$ as they seem to ...
- 101
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 ...
- 1
1
vote
1
answer
454
views
Can transformers autoregressively generate a sequence of embeddings (instead of predictions)?
Is it theoretically possible to use a transformer architecture to autoregressively generate a sequence of embedding vectors, instead of discrete tokens?
For example, if I were to provide an input of a ...
- 11
1
vote
2
answers
945
views
How is the padding mask incorporated in the attention formula?
I have been looking for the answer in other questions but no one tackled that. I want to ask you how is the padding mask considered in the formula of attention?
The attention formula taking into ...
- 575
2
votes
1
answer
107
views
Is the problem of Language Modelling a Well-Posed Learning Problem?
Hadamard defines (Well-posed problem (Wikipedia)) a well-posed problem as one for which:
a solution exists,
the solution is unique,
the solution depends continuously on the data (e.g. it is stable)
...
1
vote
0
answers
51
views
The model's accuracy becomes suddenly so unreasonably good at beginning of the training process. I need an explaination
I am practicing machine translation using seq2seq model (more specifically with GRU/LSTM units). The following is my first model:
This model first archived about 0.03 accuracy score and gradually ...
- 11
4
votes
1
answer
983
views
Difference between dot product attention and "matrix attention"
As far as I know, attention was first introduced in Learning To Align And Translate.
There, the core mechanism which is able to disregard the sequence length, is a dynamically-built matrix, of shape ...
- 787
-1
votes
1
answer
210
views
Understanding self attention - How come there is no connection between different states?
During trying to understand transformers by reading Attention is all you need, I noticed the authors constantly refer to "self attention" without explaining it.
The original attention ...
- 787
0
votes
0
answers
273
views
Increasing "output_sequence_length" in TextVectorization layer worsens model's performance
When exploring the Twitter Sentiment Analysis dataset on Kaggle, I came up with a model that looks like this:
...
- 167
1
vote
1
answer
72
views
Many To One LSTM - Can I Use the Same Sequence as Input from Previous Timesteps?
I'm new to LSTMs, and I'm trying to do a basic timeseries prediction using stock prices. However, I'm a bit confused as to how the LSTM is supposed to remember outputs from previous timesteps when it ...
- 111
0
votes
2
answers
85
views
Which model should I apply on sequential data?
I need to predict a binary vector given a sequential dataset meaning the current datapoint depends on its predecessors as well as (known) successors.
So, it looks something like this:
Given the ...
- 101
0
votes
1
answer
98
views
Limitations of LSTMs
I'm training an LSTM model for classification on accelerometer data, and I get better results when I downsample the signal to 25 Hz than when I use a 50 Hz signal.
I use the same time frame of 1.5 ...
- 121
1
vote
1
answer
60
views
Make an NN utilize other NNs as part of its decision process
Suppose I have a NN that learns to predict the time it takes a robot to move between two jobs. That's three inputs (for starters): robot, job A, job B. Not all robots travel at the same speed, and ...
- 111
0
votes
1
answer
650
views
1D Sequence Classification with self-supervised learning
I am working on a multi-class classification task on long one-dimensional sequences. The sequence length may vary in the range $[512, 30720]$, and there is one feature associated each time-step in the ...
- 101
1
vote
2
answers
275
views
Is there any reason for giving an index to a token based on its frequency in the text?
In pre-processing of text, we need to assign a number for each token in a text. Then only we can pass it to a model. In pre-processing of text, we need to assign a number for each token in a text. The ...
- 3,990
0
votes
1
answer
53
views
What does it mean by "dynamics of a sequence" mathematically?
Consider the following paragraph from the topic named sequential models from the textbook titled Dive into Deep Learning
Both cases raise the obvious question of how to generate training
data. One ...
- 3,990
2
votes
1
answer
2k
views
Sequence Embedding using embedding layer: how does the network architecture influence it?
I want to obtain a dense vector representation of protein sequences so that I can meaningfully represent them in an embedding space. We can consider them as sequences of letters, in particular there ...
1
vote
0
answers
178
views
How to compute the loss for a sequence labeling task without the Softmax distribution?
For a sequence labeling task (NER), we compute the loss by passing the softmax distribution of the classes (e.g. vocabulary) with the gold label to the loss function (...
- 123
6
votes
3
answers
3k
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 ...
- 2,998
8
votes
4
answers
18k
views
How can I predict the next number in a non-obvious sequence?
I've got an array of integers ranging from -3 to +3.
Example: [1, 3, -2, 0, 0, 1]
The array has no obvious pattern since it represents bipolar disorder mood swings.
What is the most suitable approach ...
- 183
1
vote
0
answers
181
views
Couldn't the self-attention mechanism be replaced with a global depth-wise convolution?
The main advantages of the self-attention mechanism are:
Ability to capture long-range dependencies
Ease to parallelize on GPU or TPU
However, I wonder why the same goals cannot be achieved by ...
- 2,754
3
votes
1
answer
691
views
Is seq2seq the best model when input/output sequences have fixed length?
I understand that seq2seq models are perfectly suitable when the input and/or the output have variable lengths. However, if we know exactly the input/output sequence lengths of the neural network. Is ...
- 31
4
votes
1
answer
3k
views
What is the proper way to process continuous sequence data, such as time-series, using the Transformer?
What is the right way to input continuous, temporal (time-series) data into the Transformer? Assume we're using the basic TransformerBlock here.
Since data is continuous with no tokens, Token ...
- 316
0
votes
0
answers
248
views
What is the difference between model setup, model configuration, and model customization?
In the context of research papers related to deep learning models, the authors usually mention these terms in the experiment section when they are talking about the model: configuration, setup. For ...
- 103
4
votes
1
answer
171
views
Can Reinforcement Learning be used to generate sequences?
Can we use reinforcement learning for sequence-to-sequence tasks? If yes, whether or not this is a good choice, how could this be done?
2
votes
0
answers
57
views
Are there any successful applications of transformers of small size (<10k weights)?
In the problems of NLP and sequence modeling, the Transformer architectures based on the self-attention mechanism (proposed in Attention Is All You Need) have achieved impressive results and now are ...
- 2,754
0
votes
1
answer
36
views
Normalization of possibly not fully representative data
I am trying to train a classification RNN model on a sequence of table medical data, but I stuck with the normalization problem. I realized that I cannot simply use MinMaxScaler, because of 3 problems:...
- 111
1
vote
0
answers
71
views
Representing variable-length sequences
I want to train a model over variable-length sequential data (e.g. the temperature at different times of day) where the output depends on what the temperature is at a time ...
- 111
1
vote
0
answers
44
views
Building a resume recommendation for a job post?
There are few challenges I am facing when building a resume recommendation for a particular job positing.
Let's say we convert the resume into a vector on n-dimensions and job description also as an n-...
- 149
1
vote
1
answer
1k
views
Can we use transformers for audio classification tasks?
Since transformers are good at processing sequential data, can we also use them for audio classification problems (same as RNNs)?
2
votes
1
answer
280
views
Mining repeated subsequences in a given sequence
Given an alphabet $I=\left\{i_1,i_2,\dots,i_n\right\}$ and a sequence $S=[e_1,e_2,\dots,e_m]$, where items $e_j \in I$, I am interested in finding every single pattern (subsequence of $S$) that ...
- 73
2
votes
0
answers
148
views
NLP Bible verse division problem: Whats the best model/method?
I'm working on a project compiling various versions of the Bible into a dataset. For the most part versions separate verses discreetly. In some versions, however, verses are combined. Instead of verse ...
- 121
1
vote
0
answers
40
views
What's the difference between RNNs and Feed Forward Neural Networks if a fixed size vector can preserve sequential information?
I was watching a Youtube video in which the problem of trying to predict the last word in a sentence was posed. The sentence was "I took my cat for a" and the last word was "walk"....
0
votes
0
answers
37
views
Network design to learn multiple sequences of multiple categories
For learning a single sequence, LSTM only should suffice.
However, my situation is different here. I have a list of sequences to learn:
The sale volumes of 12 months, these are the sequences
And ...
- 1,318
0
votes
1
answer
1k
views
Number of LSTM layers needed to learn a certain number of sequences
Theoretically, number of units for a LSTM layer is the number of hidden states or the max length of sequences as per my practice.
For example, in Keras:
...
- 1,318
0
votes
1
answer
102
views
Text classification of non-equal length texts, should I pad left or right?
Text classification of equal length texts works without padding, but in reality, practically, texts never have the same length.
For example, spam filtering on blog article:
...
- 1,318
2
votes
1
answer
232
views
Do transformers have success in other domains different than NLP?
Everybody knows how successful transformers have been in NLP. Is there known work on other domains (e.g that also have a sequential natural way of occurring, such as stock price prediction or other ...
- 242
2
votes
1
answer
147
views
How do LSTM or GRU gates learn to specialize in their desired tasks?
While I was studying the equations for the computation inside GRU and LSTM units, I realized that although the different gates have different Weight matrices, their overall structure is the same. They ...
- 23
4
votes
1
answer
671
views
How can I use machine learning to predict properties (such as the area) of simple polygons?
Imagine a set of simple (non-self-intersecting) polygons given by the coordinate pairs of their vertices $[(x_1, y_1), (x_2, y_2), \dots,(x_n, y_n)]$. The polygons in the set have a different number ...
1
vote
0
answers
35
views
length independent sequence classification methods
I am looking to do sequence classification using deep learning. The length of my sequences can vary from a few hundred to several tens of thousands of characters. I was wondering what is a good ...
- 11
2
votes
1
answer
476
views
How does the number of stacked LSTM layers or units in each layer affect the model complexity?
I playing around sequence modeling to forecast the weather using LSTM.
How does the number of layers or units in each layer exactly affect the model complexity (in an LSTM)? For example, if I ...
- 121
1
vote
0
answers
70
views
Model for supervised sequence classification task
The Problem
I am currently working on a sequence classification problem I try to solve with machine learning.
The target variable is the current state of a system.
This target variable is following a ...
- 31
1
vote
0
answers
41
views
What's the best method to predict/generate signal from one sensor (source) to signal from another another (target)?
I was wondering what is the best method out there to find relationship between two 1D signals so that I can predict/generate one (source) from the other (target). For example, let's say that in ...
1
vote
0
answers
87
views
How to pad sequences during training for an encoder decoder model
I've got an encoder-decoder model for character level English language spelling correction, it is pretty basic stuff with a two LSTM encoder and another LSTM decoder.
However, up until now, I have ...
- 111
1
vote
0
answers
51
views
Generation of realistic real-valued sequences using Wasserstein GAN fails
My goal is to generate artificial sequences of real-valued data (e.g. time series) with GANs. Starting simple I tried to generate realistic sine-waves using a Wasserstein GAN. But even on this simple ...
- 111
3
votes
2
answers
763
views
How to use LSTM to generate a paragraph
A LSTM model can be trained to generate text sequences by feeding the first word. After feeding the first word, the model will generate a sequence of words (a sentence). Feed the first word to get the ...
- 1,318