Matchtigs: minimum plain text representation of kmer sets

Kmer-based methods are widely used in bioinformatics, which raises the question of what is the smallest practically usable representation (i.e. plain text) of a set of kmers. We propose a polynomial algorithm computing a minimum such representation (which was previously posed as a potentially NP-hard open problem), as well as an efficient near-minimum greedy heuristic. When compressing genomes of large model organisms, read sets thereof or bacterial pangenomes, with only a minor runtime increase, we decrease the size of the representation by up to 60% over unitigs and 27% over previous work. Additionally, the number of strings is decreased by up to 97% over unitigs and 91% over previous work. Finally we show that a small representation has advantages in downstream applications, as it speeds up queries on the popular kmer indexing tool Bifrost by 1.66x over unitigs and 1.29x over previous work.

Against Membership Inference Attack: Pruning is All You Need

The large model size, high computational operations, and vulnerability against membership inference attack (MIA) have impeded deep learning or deep neural networks (DNNs) popularity, especially on mobile devices. To address the challenge, we envision that the weight pruning technique will help DNNs against MIA while reducing model storage and computational operation. In this work, we propose a pruning algorithm, and we show that the proposed algorithm can find a subnetwork that can prevent privacy leakage from MIA and achieves competitive accuracy with the original DNNs. We also verify our theoretical insights with experiments. Our experimental results illustrate that the attack accuracy using model compression is up to 13.6% and 10% lower than that of the baseline and Min-Max game, accordingly.

Using Task Farming to Optimise a Street-Scale Resolution Air Quality Model of the West Midlands (UK)

High resolution air quality models combining emissions, chemical processes, dispersion and dynamical treatments are necessary to develop effective policies for clean air in urban environments, but can have high computational demand. We demonstrate the application of task farming to reduce runtime for ADMS-Urban, a quasi-Gaussian plume air dispersion model. The model represents the full range of source types (point, road and grid sources) occurring in an urban area at high resolution. Here, we implement and evaluate the option to automatically split up a large model domain into smaller sub-regions, each of which can then be executed concurrently on multiple cores of a HPC or across a PC network, a technique known as task farming. The approach has been tested for a large model domain covering the West Midlands, UK (902 km2), as part of modelling work in the WM-Air (West Midlands Air Quality Improvement Programme) project. Compared to the measurement data, overall, the model performs well. Air quality maps for annual/subset averages and percentiles are generated. For this air quality modelling application of task farming, the optimisation process has reduced weeks of model execution time to approximately 35 h for a single model configuration of annual calculations.

Dropout- A Detailed Survey

Deep Neural Networks are very complex and have large number of parameters. Shortlisting the parameters that influence the model prediction is not possible as each has equal significance. These neural nets have powerful learning skills can model training data well enough. However, in most of these conditions, the models are over-fitting. Combining predictions from large neural nets where neurons are co-dependent alters the performance of the model. Dropout addresses the problem of overfitting and slow convergence in deep neural nets. The core concept of dropout technique is to randomly drop units and their connections from the neural network during training phase. This prevents units from co-adapting and thus improving the performance. The central mechanism behind dropout is to take a large model that overfits easily and repeatedly sample and train smaller sub-models from it. This paper provides an introduction to dropout, the history behind its design and various dropout methods.

EfficientPose: Efficient human pose estimation with neural architecture search

Human pose estimation from image and video is a key task in many multimedia applications. Previous methods achieve great performance but rarely take efficiency into consideration, which makes it difficult to implement the networks on lightweight devices. Nowadays, real-time multimedia applications call for more efficient models for better interaction. Moreover, most deep neural networks for pose estimation directly reuse networks designed for image classification as the backbone, which are not optimized for the pose estimation task. In this paper, we propose an efficient framework for human pose estimation with two parts, an efficient backbone and an efficient head. By implementing a differentiable neural architecture search method, we customize the backbone network design for pose estimation, and reduce computational cost with negligible accuracy degradation. For the efficient head, we slim the transposed convolutions and propose a spatial information correction module to promote the performance of the final prediction. In experiments, we evaluate our networks on the MPII and COCO datasets. Our smallest model requires only 0.65 GFLOPs with 88.1% [email protected] on MPII and our large model needs only 2 GFLOPs while its accuracy is competitive with the state-of-the-art large model, HRNet, which takes 9.5 GFLOPs.

Event Classification with Multi-step Machine Learning

The usefulness and value of Multi-step Machine Learning (ML), where a task is organized into connected sub-tasks with known intermediate inference goals, as opposed to a single large model learned end-to-end without intermediate sub-tasks, is presented. Pre-optimized ML models are connected and better performance is obtained by re-optimizing the connected one. The selection of an ML model from several small ML model candidates for each sub-task has been performed by using the idea based on Neural Architecture Search (NAS). In this paper, Differentiable Architecture Search (DARTS) and Single Path One-Shot NAS (SPOS-NAS) are tested, where the construction of loss functions is improved to keep all ML models smoothly learning. Using DARTS and SPOS-NAS as an optimization and selection as well as the connections for multi-step machine learning systems, we find that (1) such a system can quickly and successfully select highly performant model combinations, and (2) the selected models are consistent with baseline algorithms, such as grid search, and their outputs are well controlled.