23 Questions 20 Answers 0 Followers
Questions related from Masab Ahmad
Assume a memory access bound workload such as graph analytics, machine learning, monte-carlo simulations etc. Assume a high-end single chip GPU of the current generation (2020).
02 February 2020 735 3 View
Works such as Mi6, InvisiSpec, and IRONHIDE, and many other works, target secure isolation of target processes. How do these works rank and work in terms of performance, complexity, and usability?
05 May 2019 8,539 2 View
Weather modeling is a prediction problem, and most models are not entirely accurate. Has machine learning improved current weather models? What has improved and how does it improve predictions?
04 April 2019 918 2 View
Computing takes significant power, resources, and man-power. What are the environmental implications of computing in this context?
04 April 2019 2,804 3 View
Automation of various jobs is underway. What types of machine learning algorithms are used to replace human work?
04 April 2019 5,664 5 View
Bayesian inference is a machine learning model not as widely used as deep learning or regression models. Why is it not as widely used and how does it compare to highly used models?
04 April 2019 1,044 9 View
This question stems from the latest papers in quantum machine learning. https://www.technologyreview.com/s/613119/quantum-computing-should-supercharge-this-machine-learning-technique/
03 March 2019 5,856 3 View
Quantum computers are known to run at least a couple of applications better than conventional computers. When a quantum computer does come about to exist, how will applications be switched in and...
03 March 2019 3,671 0 View
As machine learning evolves into other domains, it is encompassing domains that use graph structures. What are the algorithms and open-source implementations in such cases?
03 March 2019 2,938 6 View
Quantum computers are known to perform extremely well on a limited number of problems at this time. For real applications, such as path planning and search in graph analytics, are quantum...
03 March 2019 8,509 2 View
Dijkstra's algorithms performs well sequentially. However, applications require even better parallel performance because of real-time constraints. Implementations such as SprayList and Relaxed...
03 March 2019 4,903 5 View
Quantum computers are known to perform well on a limited problem space. With fast incoming developments in their technology, are quantum computers going to be used as general purpose machines in...
03 March 2019 8,411 3 View
Nvidia has a larger market share of GPU sales. What are the reasons for this larger share?
03 March 2019 10,023 1 View
Performance prediction is required to optimally deploy workloads and inputs to a particular machine/accelerator in computing systems. Different predictors (e.g. AI predictors) come with different...
03 March 2019 9,993 3 View
Intel's SGX extensions create isolated application enclaves, which disallow information leakage and unverified access to private data. However, SGX is now known to be broken as some works have...
03 March 2019 4,034 4 View
Synchronization overheads blow up exponentially as more and more cores are deployed on a tiled mesh multicore. Synchronization costs increase as a multicore can only have a limited number of...
02 February 2019 8,907 2 View
Current parallel BFS algorithms are known to have reduced time complexity. However, such cases do not take into account synchronization costs which increase exponentially with the core count. Such...
02 February 2019 3,151 3 View
Graph algorithms such as BFS and SSSP (Bellman-Ford or Dijkstra's algorithm) generally exhibit a lack of locality. A vertex at the start of the graph may want to update an edge that exists in a...
02 February 2019 3,493 2 View
Current architectures already have accelerators integrated within them (e.g. GPGPUs). However future multicores are expected to have more and more cores, and hence more tiles as accelerators....
02 February 2019 6,366 2 View
Some workloads or even inputs perform well on GPUs, while others perform well on multicores. How do we decide which machine to buy for a generic problem base for optimal performance? Cost is NOT...
02 February 2019 605 4 View
Dijkstra's algorithm performs the best sequentially on a single CPU core. Bellman-Ford implementations and variants running on the GPU outperform this sequential Dijkstra case, as well as parallel...
02 February 2019 8,637 1 View
Synchronization and memory costs are becoming humongous bottlenecks in today's architectures. However, algorithm complexities assume these operations as constant, which are done in O(1) time. What...
02 February 2019 6,249 5 View
C/C++ show better performance than Python due to Python's higher level function calls and wrapping routines. However, Python's time-to-program is lower than C/C++ due to lower language complexity....
01 January 2019 1,341 9 View
