Embed Size (px)
Citation preview
Artificiell Intelligensförändrar allt? Eller?
Fredrik Heintz, Institutionen för Datavetenskap
Linköpings universitet
@FredrikHeintz
Image Classification
Speech Recognition
Artificial Intelligence
Interaction
• Human-AI collaboration
• Social and ethical aspects
• Multi-agent systems
Reasoning
• Inference
• Prediction
• Decision making
• Planning
Learning
• Deep learning
• Bayesian learning
• Reinforcement learning
Robotics / Cyber-PhysicalSystems
Decision Support Systems
AGI
“Artificial Intelligence is the science and engineering of making intelligent machines, especially intelligent computer programs.” John McCarthy, Stanford
Supervised learning
T. Mitchell, M. Jordan:“Most of the recent progress in machine learning involves mapping from a set of inputs to a set of outputs.”
Convolutional Neural Networks
http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/
sharpen
detect edges
Convolutional Neural Networks
http://cs231n.github.io/convolutional-networks/
Convolutional Neural Networks
http://cs231n.github.io/convolutional-networks/
Deep Neural Networks
http://www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/
Recurrent Neural Networks
Recurrent Neural Networks
https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/recurrent_neural_networks.html
https://blogs.nvidia.com/blog/2016/08/22/difference-deep-learning-training-inference-ai/
Types of Machine Learning
• Supervised learning
– Given input-output examples f(X)=Y, learn the function f().
• Unsupervised learning
– Given input examples, find patterns such as clusters
• Reinforcement learning
– Select and execute an action, get feedback, update policy (what action to do in which state).
https://www.techleer.com/articles/203-machine-learning-algorithm-backbone-of-emerging-technologies/
http://en.proft.me/2015/12/24/types-machine-learning-algorithms/
http://deeplearningskysthelimit.blogspot.se/2016/04/part-2-alphago-under-magnifying-glass.html
Ian J. Goodfellow, Jonathon Shlens, Christian Szegedy. Explaining and Harnessing Adversarial Examples. ICLR 2015https://arxiv.org/abs/1412.6572
Machine learning is still brittle…
Generative Adversarial Networks (GANs)
Kevin McGuinness. Deep Learning for Computer Vision: Generative models and adversarial training (UPC 2016). http://www.slideshare.net/xavigiro/deep-learning-for-computer-vision-generative-models-and-adversarial-training-upc-2016
AutoML – Using Machine Learning to Develop ML Applications
https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html
AutoML – Using Machine Learning to Develop ML Applications
https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html
AutoML – Using Machine Learning to Develop ML Applications
https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html
Visual reasoning• A neural network is asked to answer a question using a photo. For example: “Is there a same size
rubber thing in the picture as a yellow metal cylinder?” The question is nontrivial, and until recently, the problem was solved with an accuracy of only 68.5%.
• Deepmind: on the CLEVR dataset they reached a super-human accuracy of 95.5%.
Explainable AI
https://www.darpa.mil/program/explainable-artificial-intelligence
• Sample efficient learning
• Online learning
• Learning with guarantees
• Learning explainable models
• Statistical-relational learning
• Transfer learning
Traditional vs ML problem solving
The bigger system / picture
DataML
CodeModel
Hidden technical debt in Machine Learning Systems,
Sculley et. al. (NIPS 2015)
Why now?
• Software is eating the world
• The digital and the analog are becoming one
• Computer power and GPUs
• Massive amounts of labeled data
• Improved open-source algorithms due to increasing research inacademia and industry
• Critical mass
1997
Evolution of Self-Driving Cars
From Horse to Car
1903 1913
https://www.inmotionventures.com/movement-disrupted/
Prediction Date Comment
First dedicated lane where only cars in truly driverless mode are allowed on a public freeway. NET 2021
This is a bit like current day HOV lanes. My bet is the left most lane on 101 between SF and Silicon Valley (currently largely the domain of speeding Teslas in any case). People will have to have their hands on the wheel until the car is in the dedicated lane.
Such a dedicated lane where the cars communicate and drive with reduced spacing at higher speed than people are allowed to drive
NET 2024
First driverless "taxi" service in a major US city, with dedicated pick up and drop off points, and restrictions on weather and time of day.
NET 2022The pick up and drop off points will not be parking spots, but like bus stops they will be marked and restricted for that purpose only.
Such "taxi" services where the cars are also used with drivers at other times and with extended geography, in 10 major US cities
NET 2025A key predictor here is when the sensors get cheap enough that using the car with a driver and not using those sensors still makes economic sense.
Such "taxi" service as above in 50 of the 100 biggest US cities. NET 2028It will be a very slow start and roll out. The designated pick up and drop off points may be used by multiple vendors, with communication between them to schedule cars in and out.
Dedicated driverless package delivery vehicles in very restricted geographies of a major US city.
NET 2023The geographies will have to be where the roads are wide enough for other drivers to get around stopped vehicles.
A (profitable) parking garage where certain brands of cars can be left and picked up at the entrance and they will go park themselves in a human free environment.
NET 2023The economic incentive is much higher parking density, and it will require communication between the cars and the garage infrastructure.
A driverless "taxi" service in a major US city with arbitrary pick and drop off locations, even in a restricted geographical area.
NET 2032 This is what Uber, Lyft, and conventional taxi services can do today.
Driverless taxi services operating on all streets in Cambridgeport, MA, on Greenwich Village, NY,
NET 2035 Unless parking and human drivers are banned from those areas before then.
A major city bans parking and cars with drivers from a non-trivial portion of a city so that driverless cars have free reign in that area.
NET 2027BY 2031
This will be the starting point for a turning of the tide towards driverless cars.
The majority of US cities have the majority of their downtown under such rules. NET 2045
https://rodneybrooks.com/my-dated-predictions/
https://www.navigantresearch.com/research/navigant-research-leaderboard-automated-driving-vehicles
• Safety guarantees
• Decision making in complex situations
• Integration in transportation infrastructure
• Handle all operational environments
Wallenberg Autonomous Systems and Software Program (WASP)
Research ProgramThe best researchers in the field
Graduate SchoolAmbitious program, Industrial PhDs
Demonstrator ArenasDemonstrations with external parties
Recruitment Program Internationally competitive offers
Ten year program 1900 MSEK~190 million Euro
http://wasp-sweden.se/
Wallenberg Autonomous Systems and Software Program (WASP)
Extends WASP with 1000 MSEK for AI. Two tracksAI/ML: Machine Learning, Deep Learning,
eXplainable AI (XAI)AI/Math: Theoretical questions related
to AI in a wide senseAI4X Conferences in the spring:
Feb 12: IndustryFeb 27: Education & EntertainmentMar 13: HealthMar 27: Finance & ServicesApr 11: Society & Environment
WASP AI
http://wasp-sweden.se/ai/ai4x/
• Manipulation
• Human-robot collaboration
• Robustness and extended operation
• From single task to general purpose
https://www.tieto.se/sites/default/files/atoms/files/trelleborg_low.pdf
IoT
IBM Watson
• Extracting and leveraging semantics
• From single questions to continuous dialogue
• Online and stream reasoning
• Proactive decision support
• Combine model-based and data-driven approaches
https://spectrum.ieee.org/the-human-os/biomedical/diagnostics/stanford-algorithm-can-diagnose-pneumonia-better-than-radiologistshttps://arxiv.org/abs/1711.05225
NIH released data set with 112 120 chest X-ray images with 14 labeled diagnoses. 4 Stanford radiologists annotated 420 images for indications of pneumonia.After 1 month of training cheXNet outperformed all the radiologists.
“Weak human + machine + superior process was greater than a strong computer and, remarkably, greater than a strong human + machine with inferior process.”
Garry Kasparov
< <
Human-Robot Collaboration
Delegation
AdjustableAutonomy
Mixed-InitiativeInteraction
Delegate(A, B, task, constraints)
Delegate(GOP, UAV, task, constraints)Delegate(UAV, GOP, task, constraints)Important: Safety, security, trust, etc.
By varying the task and constraints parameters the degree of autonomy allowed can be controlled.
Patrick Doherty, Fredrik Heintz and Jonas Kvarnström. 2013.High-level Mission Specification and Planning for Collaborative Unmanned Aircraft Systems using Delegation. Unmanned Systems, 1(1):75–119. World Scientific.
Human Computing
https://www.mckinsey.com/global-themes/digital-disruption/harnessing-automation-for-a-future-that-works
Computational ThinkingDigital Competence
http://qaspire.com/2015/11/23/mindset-shifts-for-organizational-transformation/
Prediction is hard, especially about the future
September 11, 1933 September 12, 1933
Leo SzilardHypothesized the concept of a nuclear chain reaction
Digital photography
iPhone era
http://www.fpa-trends.com/article/how-manage-forecasting-risk
Recommendations
• Learn
– Competence development is essential, consultants can help but are temporary
• Do
– Find a real business problem that require prediction or classification and solve it
– Use off-the-shelf-tools
– Industrial PhD students / postdocs
• Scale up
– Business development at least as important as technology development
IJCAI-ECAI 2018 (ijcai-18.org), Stockholm, Sweden
Also Int. Conf. Machine Learning (ICML), Int. Conf. Autonomous Agents and
Multi Agent Systems (AAMAS), andInt. Conf. Case-Based Reasoning (ICCBR)will be in Stockholm July 9-15
Artificiell Intelligensförändrar allt? Eller?
Fredrik Heintz, Institutionen för Datavetenskap
Linköpings universitet
@FredrikHeintz
