Transfer learning needs to be from more relevant domain? - machine-learning

I am in search of a reference paper where I can find out that transfer learning needs to be from domain specific source model rather than using generalise model i.e., imagenet
For example Source dataset satellite/drone hyper/multi spectral images of plants and target dataset of hyper/multi spectral images of plants captured using agricultural robot
As compared to
Source dataset ImageNet model and target dataset images of plants captured using agricultural robot

Transfer learning is especially interesting for the accuracy if you don't have enough data. For example, this paper compared training with and without pretraining on imagenet. They claim that after 10k images, pretraining does not give better results but still allows to train faster.
Then if you have a small dataset, your question still holds whether you should pretrain on imagenet or on another dataset. I think the answer to this question is given in the following paragraph (the references there are probably of interest to you):
Do we need big data? Yes. But a generic large-scale, classification-level pre-training set is not ideal if we take into account the extra effort of collecting and cleaning data—the cost of collecting ImageNet has been largely ignored, but the ‘pre-training’ step in the ‘pre-training +fine-tuning’ paradigm is in fact not free when we scale out this paradigm. If the gain of large-scale classification-level pre-training becomes exponentially diminishing [44, 30], it would be more effective to collect data in the target domain.
Therefore, you also need to consider the quality of your satellite image dataset. Since it should be closer to your data than Imagenet it is probably better.

Related

How to deal with dataset of different features?

I am working to create an MLP model on a CEA Classification Dataset (Binary Classification). Each sample contains different 4 features, such as resistance and other values, each in its own range (resistance in hundreds, another in micros, etc.). I am still new to machine learning and this is the first real model to build. How can I deal with such data? I have tried feeding each sample to the neural network with a sigmoid activation function, but I am not getting accurate results. My assumption to deal with this kind of data is to scale it? If so, what are some resources which are useful to look at, since I do not quite understand when is scaling required.
Scaling your data can be an important step in building a machine-learning model, especially when working with neural networks. Scaling can help to ensure that all of the features in your dataset are on a similar scale, which can make it easier for the model to learn.
There are a few different ways to scale your data, such as normalization and standardization. Normalization is the process of scaling the data so that it has a minimum value of 0 and a maximum value of 1. Standardization is the process of scaling the data so that it has a mean of 0 and a standard deviation of 1.
When working with your CEA Classification dataset, it might be helpful to try both normalization and standardization to see which one works better for your specific dataset. You can use scikit-learn library's preprocessing functions like MinMaxScaler() and StandardScaler() for normalization and standardization respectively.
Additionally, it might be helpful to try different activation functions, such as ReLU or LeakyReLU, to see if they lead to more accurate results. Also, you can try adding more layers and neurons in your neural network to see if it improves the performance.
It's also important to remember that feature engineering, which includes the process of selecting the most important features, can be more important than scaling.

Instance Normalisation vs Batch normalisation

I understand that Batch Normalisation helps in faster training by turning the activation towards unit Gaussian distribution and thus tackling vanishing gradients problem. Batch norm acts is applied differently at training(use mean/var from each batch) and test time (use finalized running mean/var from training phase).
Instance normalisation, on the other hand, acts as contrast normalisation as mentioned in this paper https://arxiv.org/abs/1607.08022 . The authors mention that the output stylised images should be not depend on the contrast of the input content image and hence Instance normalisation helps.
But then should we not also use instance normalisation for image classification where class label should not depend on the contrast of input image. I have not seen any paper using instance normalisation in-place of batch normalisation for classification. What is the reason for that? Also, can and should batch and instance normalisation be used together. I am eager to get an intuitive as well as theoretical understanding of when to use which normalisation.
Definition
Let's begin with the strict definition of both:
Batch normalization
Instance normalization
As you can notice, they are doing the same thing, except for the number of input tensors that are normalized jointly. Batch version normalizes all images across the batch and spatial locations (in the CNN case, in the ordinary case it's different); instance version normalizes each element of the batch independently, i.e., across spatial locations only.
In other words, where batch norm computes one mean and std dev (thus making the distribution of the whole layer Gaussian), instance norm computes T of them, making each individual image distribution look Gaussian, but not jointly.
A simple analogy: during data pre-processing step, it's possible to normalize the data on per-image basis or normalize the whole data set.
Credit: the formulas are from here.
Which normalization is better?
The answer depends on the network architecture, in particular on what is done after the normalization layer. Image classification networks usually stack the feature maps together and wire them to the FC layer, which share weights across the batch (the modern way is to use the CONV layer instead of FC, but the argument still applies).
This is where the distribution nuances start to matter: the same neuron is going to receive the input from all images. If the variance across the batch is high, the gradient from the small activations will be completely suppressed by the high activations, which is exactly the problem that batch norm tries to solve. That's why it's fairly possible that per-instance normalization won't improve network convergence at all.
On the other hand, batch normalization adds extra noise to the training, because the result for a particular instance depends on the neighbor instances. As it turns out, this kind of noise may be either good and bad for the network. This is well explained in the "Weight Normalization" paper by Tim Salimans at al, which name recurrent neural networks and reinforcement learning DQNs as noise-sensitive applications. I'm not entirely sure, but I think that the same noise-sensitivity was the main issue in stylization task, which instance norm tried to fight. It would be interesting to check if weight norm performs better for this particular task.
Can you combine batch and instance normalization?
Though it makes a valid neural network, there's no practical use for it. Batch normalization noise is either helping the learning process (in this case it's preferable) or hurting it (in this case it's better to omit it). In both cases, leaving the network with one type of normalization is likely to improve the performance.
Great question and already answered nicely. Just to add: I found this visualisation From Kaiming He's Group Norm paper helpful.
Source: link to article on Medium contrasting the Norms
I wanted to add more information to this question since there are some more recent works in this area. Your intuition
use instance normalisation for image classification where class label
should not depend on the contrast of input image
is partly correct. I would say that a pig in broad daylight is still a pig when the image is taken at night or at dawn. However, this does not mean using instance normalization across the network will give you better result. Here are some reasons:
Color distribution still play a role. It is more likely to be a apple than an orange if it has a lot of red.
At later layers, you can no longer imagine instance normalization acts as contrast normalization. Class specific details will emerge in deeper layers and normalizing them by instance will hurt the model's performance greatly.
IBN-Net uses both batch normalization and instance normalization in their model. They only put instance normalization in early layers and have achieved improvement in both accuracy and ability to generalize. They have open sourced code here.
IN provide visual and appearance in-variance and BN accelerate training and preserve discriminative feature.
IN is preferred in Shallow layer(starting layer of CNN) so remove appearance variation and BN is preferred in deep layers(last CNN layer) should be reduce in order to maintain discrimination.

Regularization on Sample vs Full Dataset for Machine Learning

I have recently watched a video explaining that for Deep Learning, if you add more data, you don't need as much regularization, which sort of makes sense.
This being said, does this statement hold for "normal" Machine Learning algorithms like Random Forest for example ? And if so, when searching for the best hyper-parameters for the algorithm, in theory you should have as input dataset ( of course that gets further divided into cross validation sets etc ) as much data as you have, and not just a sample of it. This of course means a muuch longer training time, as for every combination of hyper-params you have X cross-validation sets which need to be trained and so on.
So basically, is it fair to assume that the params found for a decently size sample of your dataset are the "best" ones to use for the entire dataset or isn't it ?
Speaking from a statistician's point of view: it really depends on the quality of your estimator. If it's unbiased and low-variance, then a sample will be fine. If the variance is high, you'll want to use all the data you can.

Deep learning Training dataset with Caffe

I am a deep-learning newbie and working on creating a vehicle classifier for images using Caffe and have a 3-part question:
Are there any best practices in organizing classes for training a
CNN? i.e. number of classes and number of samples for each class?
For example, would I be better off this way:
(a) Vehicles - Car-Sedans/Car-Hatchback/Car-SUV/Truck-18-wheeler/.... (note this could mean several thousand classes), or
(b) have a higher level
model that classifies between car/truck/2-wheeler and so on...
and if car type then query the Car Model to get the car type
(sedan/hatchback etc)
How many training images per class is a typical best practice? I know there are several other variables that affect the accuracy of
the CNN, but what rough number is good to shoot for in each class?
Should it be a function of the number of classes in the model? For
example, if I have many classes in my model, should I provide more
samples per class?
How do we ensure we are not overfitting to class? Is there way to measure heterogeneity in training samples for a class?
Thanks in advance.
Well, the first choice that you mentioned corresponds to a very challenging task in computer vision community: fine-grained image classification, where you want to classify the subordinates of a base class, say Car! To get more info on this, you may see this paper.
According to the literature on image classification, classifying the high-level classes such as car/trucks would be much simpler for CNNs to learn since there may exist more discriminative features. I suggest to follow the second approach, that is classifying all types of cars vs. truck and so on.
Number of training samples is mainly proportional to the number of parameters, that is if you want to train a shallow model, much less samples are required. That also depends on your decision to fine-tune a pre-trained model or train a network from scratch. When sufficient samples are not available, you have to fine-tune a model on your task.
Wrestling with over-fitting has been always a problematic issue in machine learning and even CNNs are not free of them. Within the literature, some practical suggestions have been introduced to reduce the occurrence of over-fitting such as dropout layers and data-augmentation procedures.
May not included in your questions, but it seems that you should follow the fine-tuning procedure, that is initializing the network with pre-computed weights of a model on another task (say ILSVRC 201X) and adapt the weights according to your new task. This procedure is known as transfer learning (and sometimes domain adaptation) in community.

What's the difference between ANN, SVM and KNN classifiers?

I am doing remote sensing image classification. I am using the object-oriented method: first I segmented the image to different regions, then I extract the features from regions such as color, shape and texture. The number of all features in a region may be 30 and commonly there are 2000 regions in all, and I will choose 5 classes with 15 samples for every class.
In summary:
Sample data 1530
Test data 197530
How do I choose the proper classifier? If there are 3 classifiers (ANN, SVM, and KNN), which should I choose for better classification?
KNN is the most basic machine learning algorithm to paramtise and implement, but as alluded to by #etov, would likely be outperformed by SVM due to the small training data sizes. ANNs have been observed to be limited by insufficient training data also. However, KNN makes the least number of assumptions regarding your data, other than that accurate training data should form relatively discrete clusters. ANN and SVM are notoriously difficult to paramtise, especially if you wish to repeat the process using multiple datasets and rely upon certain assumptions, such as that your data is linearly separable (SVM).
I would also recommend the Random Forests algorithm as this is easy to implement and is relatively insensitive to training data size, but I would advise against using very small training data sizes.
The scikit-learn module contains these algorithms and is able to cope with large training data sizes, so you could increase the number of training data samples. the best way to know for sure would be to investigate them yourself, as suggested by #etov
If your "sample data" is the train set, it seems very small. I'd first suggest using more than 15 examples per class.
As said in the comments, it's best to match the algorithm to the problem, so you can simply test to see which algorithm works better. But to start with, I'd suggest SVM: it works better than KNN with small train sets, and generally easier to train then ANN, as there are less choices to make.
Have a look at below mind map
KNN: KNN performs well when sample size < 100K records, for non textual data. If accuracy is not high, immediately move to SVC ( Support Vector Classifier of SVM)
SVM: When sample size > 100K records, go for SVM with SGDClassifier.
ANN: ANN has evolved overtime and they are powerful. You can use both ANN and SVM in combination to classify images
More details are available #semanticscholar.org

Resources