![vocabulary: {A61K036, A61K038, A61K039, A61K041, A61K045}
instance: {A61K038/00, A61K038/005, A61K039/00}
{A61K038, A61K038, A61K039}
[ 0, 2, 1, 0, 0 ]
CPC Vectorization](https://image.slidesharecdn.com/adventures-data-science-180725125238/85/Adventures-in-Real-World-Data-Science-14-320.jpg)
![CPC classifier
(SVM)
text classifier
(SVM)
?
“The results are computed from the
resulting generated text…”
{A61K036, A61K038, A61K039,
A61K041, A61K045}
final classification
class 1 class 2[class 1, class 2]](https://image.slidesharecdn.com/adventures-data-science-180725125238/85/Adventures-in-Real-World-Data-Science-30-320.jpg)
![CPC classifier
(SVM)
text classifier
(SVM)
decision tree
“The results are computed from the
resulting generated text…”
{A61K036, A61K038, A61K039,
A61K041, A61K045}
final classification
class 1 class 2[class 1, class 2]
87% 76%
98%](https://image.slidesharecdn.com/adventures-data-science-180725125238/85/Adventures-in-Real-World-Data-Science-32-320.jpg)
Adventures in Real-World Data Science
- 1. Adventures in Real-World Data Science Automated Patent Classification Rollie D. Goodman TechLink Center
- 2. • Established as a technology transfer center in 1996 • Facilitates ~60% of DoD’s license agreements with industry • Helps small companies secure R&D contracts
- 3. • Established as a technology transfer center in 1996 • Facilitates ~60% of DoD’s license agreements with industry • Helps small companies secure R&D contracts
- 4. • Established as a technology transfer center in 1996 • Facilitates ~60% of DoD’s license agreements with industry • Helps small companies secure R&D contracts
- 6. training set: ~9,000 labeled patents
- 7. Patent Data Document Number US 9,832,220 Assignee US Air Force Inventors Kwiat, Luke; Kamhoua, Charles; Kwiat, Kevin Attorneys Mancini, Joseph A. Title Security Method for Allocation of Virtual Machines in a Cloud Computing Network Abstract A method for enhancing security in a cloud computing system by allocating virtual machines over hypervisors, in a cloud computing environment, in a security-aware fashion. The invention solves the cloud user risk problem by inducing a state such that, unless there is a change in the conditions under which the present invention operates, the cloud users do not gain by deviating from the allocation induced by the present invention. The invention’s methods include grouping virtual machines of similar loss potential on the same hypervisor, creating hypervisor environments of similar total loss, and implementing a risk tiered system of hypervisors based on expense factors. Publication Date 11-28-2017 CPC Classes H04L63/1441, G06F9/45558, H04L63/1408, H04L63/20
- 8. Patent Data Document Number US 9,832,220 Assignee US Air Force Inventors Kwiat, Luke; Kamhoua, Charles; Kwiat, Kevin Attorneys Mancini, Joseph A. Title Security Method for Allocation of Virtual Machines in a Cloud Computing Network Abstract A method for enhancing security in a cloud computing system by allocating virtual machines over hypervisors, in a cloud computing environment, in a security-aware fashion. The invention solves the cloud user risk problem by inducing a state such that, unless there is a change in the conditions under which the present invention operates, the cloud users do not gain by deviating from the allocation induced by the present invention. The invention’s methods include grouping virtual machines of similar loss potential on the same hypervisor, creating hypervisor environments of similar total loss, and implementing a risk tiered system of hypervisors based on expense factors. Publication Date 11-28-2017 CPC Classes H04L63/1441, G06F9/45558, H04L63/1408, H04L63/20
- 9. Patent Data Document Number US 9,832,220 Assignee US Air Force Inventors Kwiat, Luke; Kamhoua, Charles; Kwiat, Kevin Attorneys Mancini, Joseph A. Title Security Method for Allocation of Virtual Machines in a Cloud Computing Network Abstract A method for enhancing security in a cloud computing system by allocating virtual machines over hypervisors, in a cloud computing environment, in a security-aware fashion. The invention solves the cloud user risk problem by inducing a state such that, unless there is a change in the conditions under which the present invention operates, the cloud users do not gain by deviating from the allocation induced by the present invention. The invention’s methods include grouping virtual machines of similar loss potential on the same hypervisor, creating hypervisor environments of similar total loss, and implementing a risk tiered system of hypervisors based on expense factors. Publication Date 11-28-2017 CPC Classes H04L63/1441, G06F9/45558, H04L63/1408, H04L63/20
- 10. A01B33/028 A: Human necessities A01: Agriculture A01B: Machines for soil working in agriculture or industry A01B33: Tilling implements with rotary driven tools A01B33/02: …with tools on horizontal shaft transverse to direction of travel A01B33/028: …of the walk-behind type CPC Terms
- 11. A01B33/028 A: Human necessities A01: Agriculture A01B: Machines for soil working in agriculture or industry A01B33: Tilling implements with rotary driven tools A01B33/02: …with tools on horizontal shaft transverse to direction of travel A01B33/028: …of the walk-behind type CPC Terms
- 12. vocabulary: {A61K036, A61K038, A61K039, A61K041, A61K045} instance: {A61K038/00, A61K038/005, A61K039/00} CPC Vectorization
- 13. vocabulary: {A61K036, A61K038, A61K039, A61K041, A61K045} instance: {A61K038/00, A61K038/005, A61K039/00} {A61K038, A61K038, A61K039} CPC Vectorization
- 14. vocabulary: {A61K036, A61K038, A61K039, A61K041, A61K045} instance: {A61K038/00, A61K038/005, A61K039/00} {A61K038, A61K038, A61K039} [ 0, 2, 1, 0, 0 ] CPC Vectorization
- 20. Cross-Validation experiment 1 experiment 2 experiment 3 experiment 4 experiment 5 fold 1 fold 2 fold 3 fold 4 fold 5 overall accuracy randomized training data
- 21. Ensemble Learners • Train and combine multiple learners to solve a single problem • also: “multiple classifier systems” • Often outperform single classifiers • e.g. Netflix Competition, KDD 2009, and Kaggle
- 22. Text Processing
- 23. Text Processing • Stopwords: remove words that appear frequently but do not give any information about content • a, an, and, for, from, is, it, the, to, with…
- 24. Text Processing • Stopwords: remove words that appear frequently but do not give any information about content • a, an, and, for, from, is, it, the, to, with… • Stemming: reduce derived words to root (“stemmed”) form • different, differently, differ, differing, differed → differ
- 25. Text Processing • Stopwords: remove words that appear frequently but do not give any information about content • a, an, and, for, from, is, it, the, to, with… • Stemming: reduce derived words to root (“stemmed”) form • different, differently, differ, differing, differed → differ • Weighting: term frequency – inverse document frequency !"#$"% = '()* +)(,-(./0% ∗ log .-*5() 6+ 76/-*(.'8 .-*5() 6+ 76/-*(.'8 9ℎ()( '()* ; 6//-)8
- 26. the results are computed from the resulting generated text Text Processing
- 27. the results are computed from the resulting generated text results computed resulting generated text Text Processing
- 28. the results are computed from the resulting generated text results computed resulting generated text result comput result gener text Text Processing
- 29. the results are computed from the resulting generated text results computed resulting generated text result comput result gener text 3.03, 1.24, 0.68, 4.79. . . Text Processing
- 30. CPC classifier (SVM) text classifier (SVM) ? “The results are computed from the resulting generated text…” {A61K036, A61K038, A61K039, A61K041, A61K045} final classification class 1 class 2[class 1, class 2]
- 31. Decision Trees outlook humidity wind N Y Y Y N high low sunny overcast rainy high low outlook: {sunny, overcast, rainy} humidity: {high, low} wind: {high, low} hiking: {Yes, No}
- 32. CPC classifier (SVM) text classifier (SVM) decision tree “The results are computed from the resulting generated text…” {A61K036, A61K038, A61K039, A61K041, A61K045} final classification class 1 class 2[class 1, class 2] 87% 76% 98%
- 37. Questions?