IBM claims that its new Snap Machine learning library is 46x faster than TensorFlow.
IBM’s Snap Machine Learning (Snap ML) is an efficient, scalable machine-learning library that enables very fast training of generalized linear models.
IBM demonstrated that this new library can eliminate the training time as a bottleneck for machine-learning workloads, paving the way to a range of new applications. Much recently, the Snap ML library set a new benchmark by outperforming training time for ML models 46 times faster than TensorFlow.
Source : IBM Research
By using the online advertising dataset released by Criteo Labs, which includes more than 4 billion training examples, IBM trained a logistic regression classifier in 91.5 seconds. Prior to this, the best result for training the same model was bagged by TensorFlow, which trained the same model in 70 minutes on Google Cloud Platform.
Snap ML library,
- allows more agile development
- faster and more fine-grained exploration of the hyper-parameter space
- enables scaling to massive datasets
- makes frequent retraining of models possible in order to adapt to events as they occur
Source: Snap Machine Learning Research paper
Let’s take a look at the three distinct features of Snap ML library:
- Distributed training: This feature helps in building a data-parallel framework. This allows one to scale out and train on massive datasets that exceed the memory capacity of a single machine, which is crucial for large-scale applications.
- GPU acceleration: Implementation of specialized solvers designed to leverage the massively parallel architecture of GPUs while respecting the data locality in GPU memory in order to avoid large data transfer overheads.
- Sparse data structures: Many machine learning datasets are sparse, therefore some new optimizations have been enrolled for the algorithms used in IBM’s own system when applied to sparse data structures.
Read more about this exciting news in detail on IBM Research.