Seminars
Existing methods such as t-tests fails to efficiently compare more than two populations. ANOVA efficiently compares two or more populations. The tool is also used to test hypothesis. The seminar talk about few concepts - Hypothesis Testing, Null Hypothesis Test, Confidence Interval and F-Ratio before diving deep into the theory of ANOVA. A short introduction is given to types of ANOVA, One-Way ANOVA, Two-Way ANOVA and MANOVA while One-Way ANOVA is explained with an elucidated example.
Lack of support for inter-GPU communication from existing Training libraries and the modifications required to enable the feature being significant and complicated led to the inception of horovod. It employs efficient inter-GPU Communication using ring-reduce with construct of higher level APIs. The presentation focuses on the Parallelism Techniques, Parameter-Server Architecture and issues with Distributed Tensorflow. Later the focus shifts towards the solution - Execution of Ring-Reduce Algorithm with the use of Message-Passing-Interface.
