Knowledge Graphs and Milestone
Download as PPTX, PDF1 like384 views
This document summarizes a presentation given by Dr. Barry Norton on knowledge graphs for data fusion. Some key points discussed include: - Knowledge graphs can integrate data from various sources like video analytics, access control, sensors and background information to analyze related events. - Milestone's video management software has the capability to recognize individuals across camera streams and correlate suspicious access control events with later cybersecurity incidents using a knowledge graph approach. - The presentation discusses the history and applications of knowledge graphs, highlighting how they can provide benefits for security, transportation and other use cases when combined with video and sensor data from an Internet of Things environment.
![Erdős Numbers
“Paul Erdős (1913–1996) was an influential Hungarian mathematician who in the latter part
of his life spent a great deal of time writing papers with a large number of colleagues […]
He published more papers during his lifetime (at least 1,525) than any other
mathematician in history.
https://www.csauthors.net
https://en.wikipedia.org/wiki/Erdős_number](https://image.slidesharecdn.com/devcon19knowledgegraphs-191008062915/85/Knowledge-Graphs-and-Milestone-17-320.jpg)
Knowledge Graphs and Milestone
- 2. Dr. Barry Norton, Director of Research Knowledge Graphs for Data Fusion
- 3. Use LPR to spot an unregistered car in the car park. Recognise and track a person from their car to the door. Receive their access control event from the door and identify them. Question 1/3 We can: Why can’t we: Mail the driver of the car and ask them to register it?
- 4. Put together an access control event and a related camera stream and spot tail-gating. Question 2/3 We can: Why can’t we: Tell to which rooms & corridors the tail-gater gained access. Activate analytics and track the tail-gater. Correlate this suspicious event with a later cybersecurity event if the tail-gater gained access to IT infrastructure.
- 5. Provide traffic analytics to a city with a network of traffic cameras. Integrate traffic sensors. Spot accidents and other traffic anomalies. Question 3/3 We can: Why can’t we: Provide data for the platform partners on which road has the anomaly, which roads offer alternative routes and what their loads are.
- 6. Knowledge Graphs Video & (meta)data from video analytics Access control, sensors and IoT Background knowledge: floor plans, buildings, road networks, etc.
- 7. First computer Started programming BASIC (via ROM cartridge) on my first computer, Texas Instruments TI99/4A in 1983.
- 8. Finite State Machines (Automata) Started working as a researcher (Research Assistant) with DaimlerChrysler, modelling embedded systems for automotive sector as Finite State Automata. (Literally in the last century!) https://en.m.wikipedia.org/wiki/Finite-state_machine https://commons.wikimedia.org/wiki/File:Turnstile_state_machine_colored.svg
- 9. Labelled Transition Systems Starting work on my PhD in 2000 with μEpsilon, since acquired by Bosch, on modelling embedded systems in process algebra. Behaviour(al semantics) represented as labelled transition systems (LTSs) (aka rooted labelled directed graphs…) © ‘A Theory for Flow-Oriented Software Processes’, PhD Thesis, Barry Norton, 2010
- 10. Graphs (Networks) https://en.m.wikipedia.org/wiki/Graph_(discrete_mathematics) https://commons.wikimedia.org/wiki/File:6n-graf.svg Each graph (G) consists of a set of vertices (V), and a set of edges (E), unordered pairs, connecting nodes: G = (V, E)
- 11. Directed Graphs (DiGraphs) Each digraph (G) consists of a set of vertices (V), and a set of edges (E), ordered pairs, connecting nodes: G = (V, E) 6 4 5 1 23 E = {(1, 2), (1, 5), (2, 3), (2, 5), (3, 4), (4, 5), (4, 6)}
- 12. Directed Acyclic Graphs (DAGs) Each DAG (G) consists of a set of vertices (V), and a set of edges (E), ordered pairs, connecting nodes, such that…: G = (V, E) 6 4 5 1 23 E = {(1, 2), (1, 5), (2, 3), (2, 5), (3, 4), (4, 5), (4, 6)}
- 13. Rooted DiGraphs Each digraph (G) consists of a set of vertices (V), of which one is distinguished, and a set of edges (E), ordered pairs, connecting nodes: G = (V, v0 , E) 6 4 5 1 23 E = {(1, 2), (1, 5), (2, 3), (2, 5), (3, 4), (4, 5), (4, 6)} v0 = 1
- 14. Labelled DiGraphs Each labelled digraph (G) consists of a set of vertices (V), a set of labels (L), and a set of edges (E), triples including two nodes and a label: G = (V, E) 6 4 5 1 23 E = {(1, a, 2), (1, b, 5), (2, c, 3), (2, b, 5), (3, d, 4), L = {a, b, c, d, e} (4, b, 5), (4, e, 6)} b a b d c be
- 15. LTSs and Equivalence My PhD extended the notion of observational equivalence, based on graph bisimilarity with silent actions, to a qualitatively timed setting, demonstrated compositionality, etc., and developed an algebra capturing this equivalence. © ‘A Theory for Flow-Oriented Software Processes’, PhD Thesis, Barry Norton, 2010
- 16. Kevin Bacon Numbers An actor’s Bacon number is the number of lowest number of hops through films and co- stars until a co-starring relationship with Kevin Bacon. Kevin Bacon had a lot of co-stars in the 1980s! https://oracleofbacon.org
- 17. Erdős Numbers “Paul Erdős (1913–1996) was an influential Hungarian mathematician who in the latter part of his life spent a great deal of time writing papers with a large number of colleagues […] He published more papers during his lifetime (at least 1,525) than any other mathematician in history. https://www.csauthors.net https://en.wikipedia.org/wiki/Erdős_number
- 18. PageRank The ‘Webgraph’ is the digraph of the entire Web, where vertices are labelled with URLs, and (unlabelled) directed edges represent hyperlinks. PageRank applies a well-known technique from scientometrics / citation analysis to finding authorities on the Web, and thereby achieving a useful ranking of search results. https://en.wikipedia.org/wiki/PageRank https://en.wikipedia.org/wiki/File:PageRanks-Example.jpg https://www.milestonesys.com https://www.milestonesys.com/events/ https://www.milestonesys.com/events/ events-and-webinars/Milestone-Developer-Days/ https://www.milestonesys.com/community/
- 19. Last Bit of Maths… Graphs can be represented by adjacency matrices, where 1’s represent edges. Undirected graphs are symmetrical. Through transforms and eigenvectors, useful properties are found. 6 4 5 1 23 1 2 3 4 5 6 1 0 1 0 0 1 0 2 0 0 1 0 1 0 3 0 0 0 1 0 0 4 0 0 0 0 1 1 5 0 0 0 0 0 0 6 0 0 0 0 0 0
- 20. PageRank (In all its complicated beauty) Tl;dr PageRank can be expressed in linear algebra, since graphs are just matrices and spectral methods can be applied. https://en.wikipedia.org/wiki/PageRank
- 21. Semantic Web The Semantic Web’s central idea is a webgraph where edges are labelled with relationships defined by an ontology, which also classifies resources. The lightweight application of the technologies, in not necessarily public settings (cf. Internet versus intranet) was called Linked Data, until… https://www.w3.org/TR/2014/NOTE-rdf11-primer-20140624/ Copyright © 2003-2014 W3C® (MIT, ERCIM, Keio, Beihang)
- 22. Knowledge Graphs In 2012, the Wikimedia Foundation launched WikiData, a collaboratively-edited graph data repository. In August 2018, WikiData reached one billion edits. Also in 2012 Google announced their searches were now improved by Google Knowledge Graph… and the term stuck.
- 23. “Enough with the maths and history lesson, what does this have to do with Milestone?”
- 24. Knowledge Graphs and IoT https://www.youtube.com/watch?v=ebBTdH62yLg https://www.youtube.com/watch?v=r3yMSl5NB_Q
- 25. Knowledge Graphs In 2011 I left academia to pursue research in industry and consulted for BBC, UK Parliament and the British Museum. In 2013 I became Development Manager on the ResearchSpace project and re-branded this as building a ‘cultural heritage knowledge graph’. https://www.youtube.com/watch?v=MaAv0SE7wis
- 27. Floor Plans and Access Control https://www.youtube.com/watch?v=tCE01ji6N24
- 28. Floor Plan (Here) You are here
- 30. Floor Plan (Simple with devices) https://en.m.wikipedia.org/wiki/Floor_plan hhttps://commons.wikimedia.org/wiki/File:Sample_Floorplan.jpg Entrance LivingDoor FamilyDoor1 KitchenDoor Kitchen Living Family FamilyDoor2 Perimeterac1 cam1 cam1 ac1 EntranceDoor ac2
- 31. Knowledge Graph for Security Entrance LivingDoor FamilyDoor1 KitchenDoor Kitchen Living Family FamilyDoor2 Perimeter cam1 ac1 EntranceDoorPerson2 Person1 PersonRecognitionEvent1 “2019-09-01 15:00:00” AccessControlEvent1 Card1 “2019-09-01 15:01:00” TailgatingEvent1 “2019-09-01 15:01:01”
- 32. Knowledge Graphs for Transport Before joining Milestone, I was the Head of Digital Platform for Mærsk, the world’s largest container shipping company.
- 33. Knowledge Graphs for Road Transport
- 34. Knowledge Graphs for Simulation
- 35. Caveat: This is a Research view, not a Product announcement. ONVIF metadata is too restrictive as a data model. Device-dependent metadata can be attached to streams, but the real value comes from fusion with context: - as a graph; - with all the entities (people, cars, rooms, corridors, devices); - with an ontology. So what does this mean for the Developer Community? “We build it, so you don’t have to.” Cf.
- 39. Graphs capture and feed deep learning https://arxiv.org/pdf/1802.07007.pdf
- 40. Erdős Numbers (Again) Prof. Kamal Nasrollahi joined Milestone as Head of Machine Learning in August… https://www.csauthors.net
- 41. Who is Kamal? • Professor of Computer Vision and Machine Learning at Aalborg University (AAU) • Head of Machine Learning at Milestone Systems • Worked at Securitas in Sweden • PhD, 2010 (AAU) • Working with more than 800 students. • Has supervised several PhDs and postdocs. • Published around 90 papers
- 42. What will Kamal do at Milestone? • Understand our partners’ needs from computer vision perspective. • Lead the vision on an improved platform for partners. • Strengthen the collaboration between Milestone and academic sector • Lead our machine learning activities in Milestone.
- 43. What’s next... Continue the conversation on Developer Forum