I wrote a blog on May 18th, which I have now deleted. I deleted it because I had a second thought, prompted by an email from Jagannath Rajagopal, Intraspexion's Chief Data Scientist. He questioned why I had searched the Patent Full-Text and Image Database in the United States Patent and Trademark Office (USPTO.gov) only for "deep learning" using two search categories: Issue Date and Title.
I thought, right, why not Claims? After all, claims are an invention's "metes and bounds." See https://www.uspto.gov/web/offices/pac/mpep/s2173.html (referring to "metes and bounds" 12 times).
So I searched for "deep learning" as Term 1 and "Claim(s)" as Field 1. I also searched for each year, beginning with 2013, as Term 2 and "Issue Date" as Field 2.
Subsequently I realized that "deep neural" would be an appropriate proxy for "deep neural network(s)," and would be accurate and potentially broader. It was. These searches allowed me to surface the 167 patents the USPTO has granted during the last 5 years (from 2013 to the present). After opening each patent, I could readily see the names of the companies to which the patents were assigned. I used the Quick Search modality and the ranking below was the result.
I later used the Advanced Search modality and found a variance of one patent each for the top three, but no change in ranking. The results below are as of May 29, 2018.
The Top 5 results were:
1. Google: 23 patents;
2. IBM: 18 patents;
3. Microsoft: 13 patents;
4. Siemens Healthcare GmbH: 10 patents; and
5. Intraspexion: 7 patents.
A detailed spreadsheet based on the Quick Search modality is available upon request. The spreadsheet lists the patent in numerical order by year and the names of the Assignees, and splits out the results into two columns: one where Term 1 is "deep learning" and a second column where Term 1 is "deep neural."
# # #