James Bridle (TED, 2018)

The nightmare videos of childrens’ YouTube — and what’s wrong with the internet today
James Bridle
Jul 13, 2018

Writer and artist James Bridle … From “surprise egg” reveals and the “Finger Family Song” to algorithmically created mashups of familiar cartoon characters in violent situations … our increasingly data-driven world

The code is more or less a black box

code_is_black_boxState-of-the-Art AI: Building Tomorrow’s Intelligent Systems
Peter Norvig, Director of Research, Google
EmTech Digital 2016. May 23, 2016

01:51 the code is more or less a black box, but you can look more carefully at it and figure out what is going on there with some degree

ML is 99% human work

machine_learning_2016AI for the Common Good
Oren Etzioni, Chief Executive Officer, Allen Institute for Artificial Intelligence
EmTech Digital 2016. May 23, 2016

Machine learning is 99% human work
Deep learning inputs:
– target concept
– algorithm
– neural network design
– labeled data

Why Big Tech pays poor Kenyans to teach self-driving cars
Dave Lee
3 November 2018

Office 365 Advanced eDiscovery

Office 365 Advanced eDiscovery (formerly Equivio Analytics)
Nov 17, 2015
by Microsoft Mechanics

This uses advanced text analytics and predictive coding to perform multi-dimensional analyses of data collections

identify and filter data; including training Machine Learning to search for relevant documents

several vectors

document similarity: 65%

recall: 84%

MVA: Data Science and ML

Data Science and Machine Learning Essentials
Microsoft Virtual Academy. Level 300
02 November 2015


Data Science with Microsoft SQL Server 2016 – Free eBook

Deep reinforcement learning

Human-level control through deep reinforcement learning
Nature  518, 529–533 (26 February 2015)
Volodymyr Mnih, et al.
The theory of reinforcement learning provides a normative account1, deeply rooted in psychological2 and neuroscientific3 perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems4, 5, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms3. While reinforcement learning agents have achieved some successes in a variety of domains6, 7, 8, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks9, 10, 11 to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games.

Game-playing software holds lessons for neuroscience
DeepMind computer provides new way to investigate how the brain works.
Nature. 25 February 2015


Steve Jurvetson: Deep Learning

black_boxbeyond_engineeringDeep Learning: Intelligence from Big Data
Sep 16, 2014
Stanford Graduate School of Business

A machine learning approach inspired by the human brain, Deep Learning is taking many industries by storm. Empowered by the latest generation of commodity computing, Deep Learning begins to derive significant value from Big Data. It has already radically improved the computer’s ability to recognize speech and identify objects in images, two fundamental hallmarks of human intelligence.

Industry giants such as Google, Facebook, and Baidu have acquired most of the dominant players in this space to improve their product offerings. At the same time, startup entrepreneurs are creating a new paradigm, Intelligence as a Service, by providing APIs that democratize access to Deep Learning algorithms.

Steve Jurvetson, Partner, DFJ Ventures

Adam Berenzweig, Co-founder and CTO, Clarifai
Naveen Rao, Co-founder and CEO, Nervana Systems
Elliot Turner, Founder and CEO, AlchemyAPI
Ilya Sutskever, Research Scientist, Google Brain

19:24 Subsystem inscrutability. Black box.

20:04 Beyond Engineering
Danny Hillis, The Pattern on the Stone: “The greatest achievement of our technology may well be the creation of tools that allow us to go beyond engineering–that allow us to create more than we can understand”

33:40 Convolutional Neural Networds

34:10 current deep learning model: recognizes 10,000 categories (classes).
average adult: 90,000 words (in English, not counting personal names)

1:09:00 we can propagate a signal with almost 0 loss. Biology can’t.
Will AI look like human intelligence? The underlying structure is different (substrate: silicon vs. neurons). With constraints, AI’s behavior will look however we want.



state of the art in reverse image search (01/01/2015):
the original (and only) source of the following image is placed low in results:

Deep learning

Deep Learning
Deterministic Neural Networks
Zhirong Wu

COS598C Spring 2014: Scene Understanding

MIT Technology Review: 10 Breakthrough Technologies 2013

see also:
In the beginning was the code: Juergen Schmidhuber
TEDxUHasselt, 2013


Stochastic neural networds

An approach to predicting non-deterministic neural network behavior

Non-Deterministic Learning Dynamics in Large Neural Networks due to Structural Data Bias (2000)

Nondeterministic Discretization of Weights Improves Accuracy of Neural Networks
Marcin Wojnarski


Predictors of adolescent alcohol misuse

Neuropsychosocial profiles of current and future adolescent alcohol misusers
Robert Whelan, et al.
Nature (02 July 2014)

A comprehensive account of the causes of alcohol misuse must accommodate individual differences in biology, psychology and environment, and must disentangle cause and effect.
Animal models can demonstrate the effects of neurotoxic substances; however, they provide limited insight into the psycho-social and higher cognitive factors involved in the initiation of substance use and progression to misuse.

One can search for pre-existing risk factors by testing for endophenotypic biomarkers in non-using relatives; however, these relatives may have personality or neural resilience factors that protect them from developing dependence.
A longitudinal study has potential to identify predictors of adolescent substance misuse, particularly if it can incorporate a wide range of potential causal factors, both proximal and distal, and their influence on numerous social, psychological and biological mechanisms.

Here we apply machine learning to a wide range of data from a large sample of adolescents (n = 692) to generate models of current and future adolescent alcohol misuse that incorporate brain structure and function, individual personality and cognitive differences, environmental factors (including gestational cigarette and alcohol exposure), life experiences, and candidate genes.

These models were accurate and generalized to novel data, and point to life experiences, neurobiological differences and personality as important antecedents of binge drinking.
By identifying the vulnerability factors underlying individual differences in alcohol misuse, these models shed light on the aetiology of alcohol misuse and suggest targets for prevention.