Please take over as many tasks as you can

• (Sakkas) cuSparse on n x k sparse array
• Point size may not fit in GPU ram
• Research cuda streams
• Compute squared distance with cublas
• Check if cublas blocks or needs Synchronize
• Make sure all threads are active at all times (Check block and grid size)
• (Sklikas) Reduce used registers number
• (Sklikas) Change where centers are saved in memory (Textures, constant)

• (Tsatiris) First Draft

• (Kallas) Find dataset with many dimension
• Keep the Final error metric for each implementation, because different ones mean different time to converge
• Make sure that in the final evaluation all implementations have the same tolerance, max_iters, etc...
• Make sure that the serial k-means is the simple version and doesn't do some kind of optimization

• (OK) Write a script that executes all implementations for all datasets
• (Kallas) Improve the script by keeping the size and dimensionality of each dataset so that we can understand the efficiency of the algorithm for different parametrizations
• Integrate the script with the gpu implementations of the algorithm

• (Greg) Simulated Annealing
• Boruvka
• Yin Yang Kmeans