際際滷

際際滷Share a Scribd company logo

N1 Analytics

Download as PPTX, PDF
Amit Jaiswal
Amit Jaiswal

N1 enables you to enrich your data by combining it with data from other companies, other jurisdictions or even your competitors. By using homomorphic encryption and by taking the computation to the data you can combine datasets while staying compliant with all policies. You can then learn from these combined datasets without anybody risking exposure or loss of control of their data. You no longer need to collect data in one place to use it. This means you are not exposed to the risk of securing or maintaining third party data. So you can now get insights from data that you couldn't access previously.

1 of 24
Download to read offline
HOW DOES N1 ANALYTICS WORK AND WHAT CAN IT DO?
HOW DOES IT WORK? N1 uses 3 cutting edge technologies that enable analytics on alternative data while: STAYING COMPLIANT OF POLICIES AND LEGISLATION KEEPING EVERYBODYS DATA SAFE DOING IT IN A COST EFFECTIVE MANNER
3 TECHNOLOGIES 1 HOMOMORPHIC ENCRYPTION 2 DISTRIBUTED COMPUTING 3 PRIVACY PRESERVING RECORD LINKAGE Encrypt(5) = #jdao
HOMOMORPHIC ENCRYPTION Homomorphic encryption is a form of encryption that allows computations to be carried out on encrypted data, thus generating an encrypted result which, when decrypted, matches the result of operations performed on the original unencrypted data. 5 X 3 = 15 Encrypt(5) = #jdao Encrypt(3) = #hdah #jdao X #hdah = #onda Decrypt(#onda) = 15
DISTRIBUTED COMPUTING The N1 Solution uses a form of distributed computing where the computation goes to where the datasets reside instead of moving the data around. Sensitive data Sensitive data Coordinator
Lets look at an example to understand Privacy Preserving record linkage.
Say you have a dataset which you would like to combine with data owned by another company. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS Your dataset Other companys dataset
N1 determines which columns can be used to link records Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS N1 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS
Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS and converts them to a bunch of hashes. This is done so that even N1 cannot know the details of those fields. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS N1
Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 It then uses fuzzy matching to match the hashes and find the common records.
Next, N1 rearranges the records so they line up with their corresponding rows. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1
Where records do not match up, N1 will create dummy data so neither party will know which records exist in the others dataset. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS
What you effectively get is an enhanced dataset but one where the data never leaves its premises. This means that none of the parties, including N1, can ever see each others data. We call this Privacy Preserving record linking. Name DOB Feat 1 Feat 2 Feat 3 Feat A Feat B Feat C 111111111 DDDDDD 1252 45646 6843 ASCH FIUHAL FHAIA XXXXXXXX FFFFFFF 5235 16818 85313 FUONA BCMOH ABXJO TTTTTTTT 5555555 6353 18843 873138 DGAIBS PISA PKAB 44444444 $$$$$$$ 6254 13837 38613 YNIXB BEJAB MABS Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Now that the enhanced dataset is ready, the user can then select an algorithm to start the computation.
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG N1 uses distributed computing to send the algorithm to the first dataset and start the computation.
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 then uses a number of techniques including homomorphic encryption to encrypt the results before sending them to the next dataset.
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS Here, the computation continues using the results of the previous computation and the new data.
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS The new results are then encrypted and sent back to N1.
N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS Finally the the results can be decrypted and presented back to the N1 user.
A second way that N1 can be used is to expand your data. Say you have a dataset and want to expand the number of records by combining it with another dataset. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 9153 FIUHAL 995453 Gloria U 03/10/64 8513 BCMOH 665544 Hillary T 01/12/77 7394 PISA 448521 Eve Y 05/08/85 6425 BEJAB 335678 N1
You can use N1 to match up the common columns so you can extend the number of rows. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 N1
What you effectively get is an expanded dataset but one where the data never leaves its premises. Again, none of the parties, including N1, can ever see each others data. Name DOB Feat 1 Feat 2 Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Name DOB Ft B Ft C Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678 Name DOB Feat 1B Feat 2C Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678
Name DOB Feat 1 Feat 2 Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Name DOB Ft B Ft C Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678 The algorithm can then move between the datasets to train a model which gives superior insights than one trained on limited data. N1
WHAT CAN N1 ANALYTICS DO FOR YOU?

More Related Content

N1 Analytics

  • 1. HOW DOES N1 ANALYTICS WORK AND WHAT CAN IT DO?
  • 2. HOW DOES IT WORK? N1 uses 3 cutting edge technologies that enable analytics on alternative data while: STAYING COMPLIANT OF POLICIES AND LEGISLATION KEEPING EVERYBODYS DATA SAFE DOING IT IN A COST EFFECTIVE MANNER
  • 4. HOMOMORPHIC ENCRYPTION Homomorphic encryption is a form of encryption that allows computations to be carried out on encrypted data, thus generating an encrypted result which, when decrypted, matches the result of operations performed on the original unencrypted data. 5 X 3 = 15 Encrypt(5) = #jdao Encrypt(3) = #hdah #jdao X #hdah = #onda Decrypt(#onda) = 15
  • 5. DISTRIBUTED COMPUTING The N1 Solution uses a form of distributed computing where the computation goes to where the datasets reside instead of moving the data around. Sensitive data Sensitive data Coordinator
  • 6. Lets look at an example to understand Privacy Preserving record linkage.
  • 7. Say you have a dataset which you would like to combine with data owned by another company. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS Your dataset Other companys dataset
  • 8. N1 determines which columns can be used to link records Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS N1 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS
  • 9. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 Barb Y 02/11/92 ASCH FIUHAL FHAIA Charley X 03/10/82 FUONA BCMOH ABXJO Alex Z 01/12/81 DGAIBS PISA PKAB Eve Y 05/08/85 YNIXB BEJAB MABS and converts them to a bunch of hashes. This is done so that even N1 cannot know the details of those fields. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS N1
  • 10. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 Name DOB Ft 2 Ft 3 Ft 4 XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO 111111111 DDDDDD DGAIBS PISA PKAB KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 It then uses fuzzy matching to match the hashes and find the common records.
  • 11. Next, N1 rearranges the records so they line up with their corresponding rows. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1
  • 12. Where records do not match up, N1 will create dummy data so neither party will know which records exist in the others dataset. Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 Name DOB Feat 1 Feat 2 Ft 3 111111111 DDDDDD 1252 45646 6843 XXXXXXXX FFFFFFFF 5235 16818 85313 TTTTTTTT 5555555 6353 18843 873138 44444444 $$$$$$$ 6254 13837 38613 GGGGGGG HHHHHH BBBBB NNNNN DDDDD Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS
  • 13. What you effectively get is an enhanced dataset but one where the data never leaves its premises. This means that none of the parties, including N1, can ever see each others data. We call this Privacy Preserving record linking. Name DOB Feat 1 Feat 2 Feat 3 Feat A Feat B Feat C 111111111 DDDDDD 1252 45646 6843 ASCH FIUHAL FHAIA XXXXXXXX FFFFFFF 5235 16818 85313 FUONA BCMOH ABXJO TTTTTTTT 5555555 6353 18843 873138 DGAIBS PISA PKAB 44444444 $$$$$$$ 6254 13837 38613 YNIXB BEJAB MABS Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS
  • 14. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Now that the enhanced dataset is ready, the user can then select an algorithm to start the computation.
  • 15. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG N1 uses distributed computing to send the algorithm to the first dataset and start the computation.
  • 16. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS N1 then uses a number of techniques including homomorphic encryption to encrypt the results before sending them to the next dataset.
  • 17. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS Here, the computation continues using the results of the previous computation and the new data.
  • 18. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS The new results are then encrypted and sent back to N1.
  • 19. N1 Ft 2 Ft 3 Ft 4 DOB Name 1252 45646 6843 DDDDDD 111111111 5235 16818 85313 FFFFFFF XXXXXXXX 6353 18843 873138 5555555 TTTTTTTT 6254 13837 38613 $$$$$$$ 44444444 BBBBB NNNNN DDDDD HHHHHH GGGGGGG Name DOB Ft 2 Ft 3 Ft 4 111111111 DDDDDD DGAIBS PISA PKAB XXXXXXXX FFFFFFFF ASCH FIUHAL FHAIA TTTTTTTT 5555555 FUONA BCMOH ABXJO PPPPPPPP LLLLLLLL MMMM JJJJJJJJ UUUUU KKKKKKKK CCCCCC YNIXB BEJAB MABS Finally the the results can be decrypted and presented back to the N1 user.
  • 20. A second way that N1 can be used is to expand your data. Say you have a dataset and want to expand the number of records by combining it with another dataset. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 9153 FIUHAL 995453 Gloria U 03/10/64 8513 BCMOH 665544 Hillary T 01/12/77 7394 PISA 448521 Eve Y 05/08/85 6425 BEJAB 335678 N1
  • 21. You can use N1 to match up the common columns so you can extend the number of rows. Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 Name DOB Feat 1 Feat 2 Ft 3 Alex Z 01/12/81 1252 45646 6843 Barb Y 02/11/92 5235 16818 85313 Charley X 03/10/82 6353 18843 873138 Diane W 04/09/99 6254 13837 38613 Name DOB Ft A Ft B Ft C Frida V 02/11/73 PL678 9153 995453 Gloria U 03/10/64 HBD86 8513 665544 Hillary T 01/12/77 KN639 7394 448521 Eve Y 05/08/85 PP879 6425 335678 N1
  • 22. What you effectively get is an expanded dataset but one where the data never leaves its premises. Again, none of the parties, including N1, can ever see each others data. Name DOB Feat 1 Feat 2 Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Name DOB Ft B Ft C Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678 Name DOB Feat 1B Feat 2C Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678
  • 23. Name DOB Feat 1 Feat 2 Alex Z 01/12/81 1252 45646 Barb Y 02/11/92 5235 16818 Charley X 03/10/82 6353 18843 Diane W 04/09/99 6254 13837 Name DOB Ft B Ft C Frida V 02/11/73 9153 995453 Gloria U 03/10/64 8513 665544 Hillary T 01/12/77 7394 448521 Eve Y 05/08/85 6425 335678 The algorithm can then move between the datasets to train a model which gives superior insights than one trained on limited data. N1