Search for: All records

The process of training deep learning models produces a huge amount of meta-data, including but not limited to losses, hidden feature embeddings, and gradients. Model diagnosis tools have been developed to analyze losses and feature embeddings with the aim to improve the performance of these models. However, gradients, despite carrying rich information that is potentially relevant for model interpretation and data debugging, have yet to be fully explored due to their size and complexity. Each single gradient has a size as large as the number of parameters of the neural net - often measured in the tens of millions. This makes it extremely challenging to efficiently collect, store, and analyze large numbers of gradients in these models. In this work, we develop MetaStore to fill this gap. MetaStore leverages our observation that storing certain compact intermediate results produced in the back propagation process, namely, the prefix and suffix gradients, is sufficient for the exact restoration of the original gradient. These prefix and suffix gradients are much more compact than the original gradients, thus allowing us to address the gradient collection and storage challenges. Furthermore, MetaStore features a rich set of analytics operators that allow the users to analyze the gradients for data debugging or model interpretation. Rather than first having to restore the original gradients and then run analytics on top of this decompressed view, MetaStore directly executes these operators on the compact prefix and suffix structures, making gradient-based analytics efficient and scalable. Our experiments on popular deep learning models such as VGG, BERT, and ResNet and benchmark image and text datasets demonstrate that MetaStore outperforms strong baseline methods from 4 to 678x in storage costs and from 2 to 1000x in running time. 

Local outlier techniques are known to be effective for detecting outliers in skewed data, where subsets of the data exhibit diverse distribution properties. However, existing methods are not well equipped to support modern high-velocity data streams due to the high complexity of the detection algorithms and their volatility to data updates. To tackle these shortcomings, we propose local outlier semantics that operate at an abstraction level by leveraging kernel density estimation (KDE) to effectively detect local outliers from streaming data. A strategy to continuously detect top-N KDE-based local outliers over streams is designed, called KELOS – the first linear time complexity streaming local outlier detection approach. The first innovation of KELOS is the abstract kernel center-based KDE (aKDE) strategy. aKDE accurately yet efficiently estimates the data density at each point – essential for local outlier detection. This is based on the observation that a cluster of points close to each other tend to have a similar influence on a target point’s density estimation when used as kernel centers. These points thus can be represented by one abstract kernel center. Next, the KELOS’s inlier pruning strategy early prunes points that have no chance to become top-N outliers. This empowers KELOS to skip the computation of their data density and of the outlier status for every data point. Together aKDE and the inlier pruning strategy eliminate the performance bottleneck of streaming local outlier detection. The experimental evaluation demonstrates that KELOS is up to 6 orders of magnitude faster than existing solutions, while being highly effective in detecting local outliers from streaming data.