[1811.12019] Second-order Optimization Method for Large Mini-batch: Training ResNet-50 on ImageNet in 35 Epochs

https://arxiv.org/abs/1811.12019

Comments

Post a Comment

Popular posts from this blog

Large SAP tables which are good candidates for partitioning

Image
Not all tables are good candidates for partitioning. The tool we have developed runs through multiple checks for large tables to identify the right candidate. Generally if the answer is yes to the following questions, table partitioning may be a suitable strategy for improving performance. 1. Is the database large enough? 2. Is the table large enough? 3. Do your reports and queries have distinguishable access patterns? 4. Are you experiencing slowness and performance degradation? 5. Is the skew factor for the data distribution under reasonable limits? Below are some of the large tables which have been candidates for table partitioning amongst our SAP customers. Large material master tables MARA                   Material Master Table MARC                   Plant Material Data MARO...
Read more

Partitioning in SAP HANA database

SAP HANA provides two ways of distributing data 1. Database Partitioning - different tables being distributed across several hosts (SAP Note 2044468 - FAQ:SAP HANA Partitioning) 2. Table Partitioning - splitting a column store table and distributing across several hosts (SAP Note 2081591 - FAQ: SAP HANA Table Distribution)  SAP HANA supports: 1. Hash Partitioning 2. Range Partitioning 3. Round-robin Partitioning In SAP HANA, we can create 1. Single-Level Partitioning 2. Multi-Level Partitioning The optimizing algorithm takes factors into account such as: - Balance tables equally in a way that they are spread over all the slave nodes and make optimal use of the available physical memory on each node - Ensure that tables/table partitions which are belonging to the same table groups (see tables _SYS_RT.TABLE_PLACEMENT, SYS.TABLE_GROUPS) are located on the same nodes, as they are logically frequently queried from application side via joins and if t...
Read more

Using Deep learning to integrate range partitioning into SAP systems

Image
The principle of partitioning is very simple: It means splitting a large table in smaller parts, called partitions, grouping and separating the rows of the table based on the content of one or more columns. Rows fulfilling the same criteria are placed in the same partition. Partitioning can provide tremendous benefit to a wide variety of applications by improving performance, manageability, and availability. - By limiting the amount of data to be examined or operated on, and by providing data distribution for parallel execution, partitioning provides multiple performance benefits. - Partitioning enables you to partition tables and indexes into smaller, more manageable units, providing database administrators with the ability to pursue a divide and conquer approach to data management. - Partitioned database objects provide partition independence. This characteristic of partition independence can be an important part of a high-availability strategy. For e...
Read more