0.X.2 - Computer Science in Action Talk Notes

29 Nov 2018

Computer Science in Action

Browsing the World with Computer Vision - Tushar Sharma

Blippar
Zeg.ai
- gen 3d models of shoes, furniture, etc
aspects:
- text = NLP
- finance, forecast, etc = big data
- visual = computer vision
Classification (visual)
Blipper video
- understand environment of images that are taken
- augmented reality
Key challenges
- 3D geometry
- how far away (location of camera)
- orientation
- camera not being held steady/cam quality
Lighting w/Augmented Reality
- model texturing:
  - map texture to shape
  - lighting changes appearance
- in aug real, may not match lighting of room = may look out of place
  - developing systems to try match ambient lighting
Datasets
- annotated datasets
- want variation in dataset
  - weather conditions
  - orientations
Use cases Deep Learning
- classify cars
  - parking lot analysis
    - cater products sold in shops (petrol station) to customers
  - understand efficiency of cars, emission levels
- insurance
  - atm needs physical presence
  - https://tractable.ai/
  - help fill in forms (model, etc) after taking pic of car
Supervised learning
- performance - comparison to human
- cloud API
  - like black box
  - cross platform
Architectures
- Convolutional neural network
Data challenges
- no landmarks to accuracy (%)
  - 1k -> 96.86%
  - 2k -> 97.23%
  - 3k -> 97.43%
Modular network approach
- split photo into easy objects (car, etc)
  - send car to other network for further identification (model, etc) using API
- EID-CR-11.12 ebay car dataset
More use cases
- Toy which has camera
  - child show it thing, it say what thing is
- Outdoor & indoor localisation
  - indoor maps
    - mall
    - airport
  - augmented reality maps (arrows)
    - self driving cars
  - AR city
Rendering
- realistic
- better opportunities in virtual that not possible irl:
  - sun in room
  - massive lens
Learning resources
- fast.ai
- standford course on visual learning CS 231n
- NVIDIA - fundamentals of deep learning for Computer Vision
Ethics
- important question to ask
- face recog api
  - security improve by not cloud based (local net)
- Taking jobs away?
  - too much work, not enough humans
  - make 3D things more accessible to average people
  - let artists focus on being creative

Jaimi Anderson, Cybersecurity

Digital forensics
- HMRC statistics
- what is it?
  - recovery and investigation of matewrial found in digital devidces
  - Public sector: gov
  - Private: ediscovery forensics
What crimes = digital forensics
- high tech cyber crimes
- identity theft
- traditional
  - money laundering, etc`
collection equipment
- traditional tools
- documentation tools
- dissasembly tools
  - take comps apart
- packaging + transport
  - evidence bags
- copying tools
  - cables, imaging software, etc
Evidence sources
- hard drives
  - documents
  - encrypted files
  - steganography
  - logs, config, cache
  - unallocated space
  - metadata
- RAM
- Networks
- Portable Storage
- periphals
  - photocopiers
  - modems
  - printers
- telephony
- anything with GPS, etc
procedure
- Assess
  - secure scene
  - protect perishables
  - identify connections
  - fingerprinting, interviews
  - document secure
- Acquisition
  - copy
  - validate
  - verify
  - remove
- when copy, prove exact copy - hashes
- capture network packets
- physical + logical extraction
  - physical = partition table analysis, etc
  - logical = filesystem analysis
Applications used
- FTK
- X-Ways
- NUIX
- Encase
- Exif
- Cellebrite
Deleted files
- data still present
Hidden files
- magic numbers
- bad sectors
- alternate data streams
  - NTFS
Steganography
- sometimes used to supplement encryption
- hiding stuff in images
Cases
- BTK killer
- the corocorans
- Kari Baker

A-Level Computer Science Exam Tech, Paul Long

keep up to date (ethics)
- ethical issues
- legal issues
- cultural issues
- envrionment issues
- privacy issues
- development in technology
cases
- cockroaches to find survivors after natural disaster
- paralysed person move arm by thinking abt moving it
exam tech
- highlight things in exams
- plan ans
  - mind map
- well structed essay
  - logical structre (paragraphs, subheadings)
  - cover all areas asked for
  - examples
  - buzz words
  - contextualise
  - advantages/disadvs
- pace yourself
  - prioritise questions
  - not worth spending ages on Q can’t answer
- try maximise marks not get maximum marks

Artificial Intelligence - Fact & Fiction, Nigel Shadbolt

Evolution
- making stone tools
- technology we use changing our brains
- amazing, hidden away in Compsci
- moores law
- cryden’s law
  - memory density
can now search in impossibly large spaces now
- AlphaGo
Applications
- Google knowledge graph
- surgery robotic
- self-driving car
will brush against the boundaries of philosophy, psychology
KBS (Knowledge Based Systems)
rule-based systems approach
- uncertainty logics
- OOP representations
1990s onwards, NNs
- Eliott & Shadbolt - Modelling neural networks
- neurotrophic factor
2000s Semantic Web Project
- Google knowledge graph - reading/analysing HTML
Neural Networks
- Find correlations in Data
- reinforcement learning
- Applications
- Deepmind breakout
- predict news cycle
- Diagnosis
  - retinal disease
- Drug discovery
- Read Lips - LipNet