Assume that I want to predict the next words of a text written in an app on a smartphone, similarly to Gmail smart-compose.
Now there are some constraints that do matter
the model should be small (perhaps a few MB)
it should not require to big resources
the inference time should be small, such that the predictions are useful for the end user while typing.
Of course, the predictions should be better than random, but being worse than BERT would be acceptable as tradeoff.
Could you tell me which steps you would take for making such a thing reality?
Related
I study ML and I see that most of the time the focus of the algorithms is run time and not accuracy. Reducing features, taking sample from the data set, using approximation and so on.
Im not sure why its the focus since once I trained my model I dont need to train it anymore if my accuracy is high enough and for that if it will take me 1 hours or 10 days to train my model it does not really matter because I do it only 1 time and my goal is to predict as better as I can my outcomes (minimum loss).
If I train a model to differ between cats and dogs I want it to be the most accurate it can be and not the fasted since once I trained this model I dont need to train any more models.
I can understand why models that depends on fasting changing data need this focus of speed but for general training models I dont understand why the focus is on speed.
Speed is relative term. Accuracy is also relative depending on the difficulty of the task. Currently the goal is to achieve human-like performance for application at reasonable costs because this will replace human labor and cut costs.
From what I have seen in reading papers, people usually focus on accuracy first to produce something that works. Then do ablation studies - studies where pieces of the models are removed or modified - to achieve the same performance in less time or memory requirements.
The field is very experimentally validated. There really isn't much of a theory that states why CNN work so well other than that it can model any function given non-linear activations functions. (https://en.wikipedia.org/wiki/Universal_approximation_theorem) There have been some recent efforts to explain why it works well. One I recall is MobileNetV2: Inverted Residuals and Linear Bottlenecks. The explaination of embedding data into a low dimensional space without losing information might be worth reading.
I have images of unique products that are used at my workplace. I can't imagine that the inception database already has similar items that it has been trained on.
I tried to train a model using YOLO. It was taking a very very long time. Maybe 7minutes between epochs; and I wanted to do 1000 epochs due to small data size.
I used tiny-yolov2-voc cfg/weight on 1.0 GPU. I had a video of the item but i broke it up into frames so i could annotate. I then attempted to train on the images (not video). The products are healthcare related. Basically anything that a hospital would use.
Ive also used the inception method on images I got from Google. I noticed that inception method was very fast and resulted in accurate predictions. However, i'm worried that my images are too unique for inception to work.
Which method is best to use?
If you recommend YOLO, can you please provide suggestions on how to speed up the training phase?
If you recommend inception, can you please provide an explanation why it would work on unique images? I guess i'm having trouble understanding how inception knows which item i'm trying to train on without me providing annotations.
Thanks in advance
Just my impression (no recommendation or even related experience)
Having a look at the Hardware recommendations related to darknet a assumption is that you might stock up your own hardware to get faster results.
I read about the currently three different versions of YOLO and expect there are lot's of GFLOPS training included if you download the recommended files, but if the models never fit to your products then for you they never might be very helpful.
I must admit I've neither been active with YOLO nor with Tensorflow, so my impression might not be helpful at all.
If you see some videos of YOLO you can remark that sometimes a camel is labeled with horse and the accuracy seems being bad but it depends on the threshold that is applied to the images, so the videos look amazing as it seems the recognition is done so fast but with higher accuracy the process would slow down - also depending on the trained motives.
They never hide it though, they explain on an image where a dog is labeled as cow and a horse as sheep (Version 2) that in combination with darknet it's getting much faster but less accurate too, so usage of darknet is an important aspect too.
The information about details seems being quite bad on the websites of YOLO, they present it more like you'd do with a popstar, in comparison the website of Tensorflow looks more academic and is informing about the mathematics behind the framework.
Concerning Tensorflow I don't know about the hardware-recommendations, but as you wrote your results are useful, probably they are a bit or even much less.
My impression is that YOLO is primary intended for real-time detection in (live-)videos and needs much training for high accuracy. So depending on your use-case it might be right but you'd to invest in hardware probably for professional usage.
This is not an opinion against Tensorflow but that I had to verify more and it seems taking more time to get an impression. Concerning Tensorflow in the moment I even can't say if it can be used for real-time-detection, how accurate it is then and if the results are then still better then those of YOLO.
My assumption is that concerning both solutions it's a matter of involved elements (like the decision if to include darknet for speed), configuration, training and adjustments. Probably there is always something to increase in speed and accuracy, so investing in a system for recognition won't be static process with fixed end in timeline, but a steady process.
This is just a short overview of my impressions, I've never any experience with any recognition-software and hardly recommend that you make any decision based on my words.
Just if you want to do use any recognition software professional, especially for real-time-recognition, then you've to invest in hardware probably.
To my understanding of your problem you need you need inception with the capability of identifying your unique images. In this circumstance you can use transfer-learning on the inception model. With transfer-learning you can still train inception your own pictures while retaining the previous knowledge of inception.
More on transfer-learning
I have a use case where I have around 100 images each of 10000 unique items. I have 10 items with me which are all from the 10000 set and I know which 10 items too but only at the time of testing on live data. I have to now match the 10 items with their names. What would be an efficient way to recognise these items? I have full control of training environment background and the testing environment background. If I make one model of all 10000 items, will it scale? Or should I make 10000 different models and run the 10 items on the 10 models I have pretrained.
Your question is regarding something called "one-vs-all classification" you can do a google search for that, the first hit is a video lecture by Andrew Ng that's almost certainly worth watching.
The question has been long studied and in a plethora of contexts. The answer to your question does very much depend on what model you use. But I'll assume that, if you're doing image classification, you are using convolutional neural networks, because, after all, they're state of the art for most such image classification tasks.
In the context of convolutional networks, there is something called "Multi task learning" that you should read up on. Boiled down to a single sentence, the concept is that the more you ask the network to learn the better it is at the individual tasks. So, in this case, you're almost certain to perform better training 1 model on 10,000 classes than 10,000 classes each performing a one-vs-all classification scheme.
Take for example the 1,000 class Imagenet dataset, and CIFAR-10's 10 class dataset. It has been demonstrated in numerous papers that first training against Imagenet's 1,000 class dataset, and then simply replacing the last layer with a 10 class output and re-training on CIFAR-10's dataset will produce a better result than just training on CIFAR-10's dataset alone. There are admittedly multiple reasons for this result, Imagenet is a larger dataset. But the richness of class labels, multi-task learning, in the Imagenet dataset is certainly among the reasons for this result.
So that was a long winded way of saying, use one model with 10,000 classes.
An aside:
If you want to get really, really interesting, and jump into the realm of research level thinking, you might consider a 1-hot vector of 10,000 classes rather sparse and start thinking about whether you could reduce the dimensionality of your output layer using an embedding. An embedding would be a dense vector, let's say size 100 as a good starting point. Now class labels turn into clusters of points in your 100 dimensional space. I bet your network will perform even better under these conditions.
If this little aside didn't make sense, it's completely safe to ignore it, your 10,000 class output is fine. But if it did peek your interest look up information on Word2Vec, and read this really nice post on how face recognition is achieved using embeddings: https://medium.com/#ageitgey/machine-learning-is-fun-part-4-modern-face-recognition-with-deep-learning-c3cffc121d78. You might also consider using an Auto Encoder to generate an embedding for the images (though I favor triplet embeddings as typically used in face recognition myself).
When training detection models, are images that are used in real life better (i.e. higher accuracy / mAP) than images of the same object but in the form of stock photo?
The more variety the better. If you train a network on images that all have a white background and expect it to perform under conditions with noisy backgrounds you should expect the results on unseen data to perform worse because the network never had a chance to learn distinguiting features of target object vs. background objects.
If you have images with transparent backgrounds one form of data augmentation that would be expected to improve results would be to place that image against many random backgrounds. The closer you come to realistic renderings of an image the better you can expect your results to be.
The more realistic examples you can augment your training dataset with, the better. Note that it generally does not help to add random noise to your data to generate larger training datasets, it only improves results when your expanded dataset contains realistic variants of the original images in the dataset.
My motto when training neural networks is this: The network will cheat any chance it gets. It will learn impressively well, but given the opportunity, it will take shortcuts. Don't let it take shortcuts. That often translates to: Make the problem harder such that no shortcut exists for it to take. Neural networks often perform better under more difficult conditions because the simplest solution it can arrive at is also the most general purpose. Read up on multi-task learning for some exciting examples that provide great food-for-thought.
I am learning about Word2Vec using the TensorFlow tutorial. The code I am running for Word2Vec is also from the TensorFlow tutorial: https://github.com/tensorflow/models/blob/master/tutorials/embedding/word2vec_optimized.py . When I ran the code for 15 epochs, the test accuracy was around 30%. When I ran for 100 epochs, test accuracy got up to around 39%. I am using the Text8 dataset for training and questions-words.txt for evaluation.
Do I need to run for more epochs? Should I be using a different dataset? How can I improve test accuracy?
Larger datasets are better; text8 is very, very small – sufficient for showing some of the analogy-solving power of word-vectors, but not good enough for other purposes.
More iterations may help squeeze slightly stronger vectors out of smaller datasets, but with diminishing returns. (No number of extra iterations over a weak dataset can extract the same rich interrelationships that a larger, more varied corpus can provide.)
There's a related text9 from the same source that if I recall correctly, is 10x larger. You'll likely get better evaluation results from using it, than from doing 10x more iterations on text8.
I believe the 3 million pretrained vectors Google once released – the GoogleNews set – were trained on a corpus of 100 billion words' worth of news articles, but with just 3 passes.
Note that there's no single standard for word-vector quality: the questions-words.txt analogy solving is just one convenient evaluation, but it's possible the word-vectors best at that won't be best at your own domain-specific analyses. Similarly, word-vectors trained on one domain of text, like the GoogleNews set from news articles, might underperform compared to text that better matches your domain (like perhaps forum posts, scientific articles, etc. – which all use different words in different ways).
So it's often best to use your own corpus, and your own goal-specific quantitative evaluation, to help adjust corpus/parameter choices.