Amreek Singh

A novel approach to accelerate calibration process of a k-nearest neighbours classifier using GPU

General purpose data parallel computing with graphical processing unit (GPU) is much structured today with NVIDIA CUDA and other parallel programming frameworks. Exploiting the CUDA programming framework, the present work proposes a novel methodology formulated around the GPU hardware architecture and memory hierarchy to accelerate the calibration process of a classification model named eNN10. Primarily developed for avalanche forecasting, eNN10 is based on brute force k-nearest neighbours (k-NN) approach and employs snow-meteorological variables to search for past days with similar conditions. The events associated with past similar days are then analysed to generate forecast. The model is required to be calibrated regularly to ensure higher degree of forecast accuracy in terms of Heidke skill score (HSS). The calibration of eNN10 is carried out by Artificial Bee Colony (ABC) algorithm, a swarm intelligence driven population based metaheuristic algorithm, and it requires thousands of HSS evaluations during the complete calibration process. A MATLAB sequential code for calibration runs for over 400 minutes and the proposed methodology delivered about 10 acceleration in calibration process. The methodology combines primitives of parallel implementations of brute force k-NN algorithm with that of population based metaheuristic algorithms and is scalable to deal with other similar real-world problems. The major objective of this paper is to highlight the methodology and associated future research areas. The SIMD model of parallel computing fitted into calibration sub-processes.A methodology formulated around GPU hardware architecture and memory hierarchy.Combines primitives of parallel implementations of ABC algorithm and k-NN algorithm.NVIDIA Tesla C2050 GPU used with CUDA programming framework.Over 10 acceleration achieved in calibration process.

Site-specific analog weather-forecast system for northwest Himalaya, India

Abstract In an analog weather-forecasting procedure, recorded weather in the past analogs corresponding to the current weather situation is used to predict future weather. Consistent with the procedure, a theoretical framework is developed to predict weather at a specific site in the Pir Panjal range of the northwest Himalaya, India, using surface weather observations of the past ten winters (1991/92 to 2001/02) 3 days in advance. Weather predictions were made as snow day with quantitative snowfall category or no-snow day, for day1 through day3. As currently deployed, the procedure routinely provides a 3 day point weather forecast as guidance information to a weather and avalanche forecaster. Forecasts by analog model are evaluated by the various accuracy measures achieved for an independent dataset of three winters (2002/03 to 2004/05). The results indicate that weather forecasts by analog model are quite reliable, in that forecast accuracy corresponds closely to the relative frequencies of observed weather events. Moreover, qualitative weather (snow day or no-snow day) and quantitative categorical snowfall forecasts (quantitative snowfall category for snow day) are better than reference forecasts based on persistence and climatology for day1 predictions. Site-specific snowfall forecast guidance may play a major role in assessing avalanche danger, and accordingly formulating an avalanche forecast for a given area in advance.