deep-learning

Neural Networks Are Easily Fooled by Strange Poses of Familiar Objects

There is a paper recently published to show the Neural Networks are never perfect and being easily fooled.  In the paper, the research team is giving “Strange Poses of Familiar Objects” for a deep neural network engine to recognize.  However, the result of detection is extremely bad with more than 99% the poses are mis-classified.

Here is the original paper published:

https://arxiv.org/pdf/1811.11553.pdf

In reality, the collection of data may lead to “Strange Poses” and not a perfect data.  Thus, the writers of the paper suggested 3D datasets / objects to study rather than just 2D image to reduce the chance of classification failures.

IBM - Modern Data Warehouse Example

Let’s see top management viewpoint on modern data warehouse

Lots of technologists are promoting replacement of data warehouse by data lake.  In the view point of our data science team, a modernized data warehouse should carry its own value for the analytics nowadays.

There is an article by TDWI.org sharing the viewpoint of CEO of Yellowbrick – Neil Carson what vital technology and tools being used.

Original Article
https://tdwi.org/articles/2019/02/08/dwt-all-ceo-qa-modern-data-warehouse-and-analytics-success.aspx

His opinion is aligning with our data science evangelist – Samuel Sum on Hadoop.  It is a large-scale data store but it is not easy to access and manage in many situations.  It is always better to have a structured data store like data warehouse for easy user access.  Also, a key-point of a successful data analytic environment is the storage speed storing the data.  With SSD (flash memory), the data warehouse (both ETL and access) are now several times faster than before.

Finally, the editor of this webpage suggests read an article of Samuel Sum – talking about the Data Lake (Data Lake VS Data Warehouse).

IBM - Modern Data Warehouse Example