Slides bij meetup https://www.meetup.com/DigAnt-Cafe/events/247417961/

1. LIMAI⇾∞= HR
BERNIE CAESSENS | MANAGING PARTNER
AI & HR MEETUP | DIGIPOLIS | FEB 2018

2. INDUSTRIE 4.0: HET 2E MACHINETIJDPERK
INTEROPERABILITEIT
• Het vermogen van machines en mensen om te connecteren en te
communiceren
TRANSPARANTIE
• De mogelijkheid van informatiesystemen om een digitale (virtuele) kopij van de
werkelijkheid te maken
ONDERSTEUNING
• De mogelijkheid van machines om mensen te ondersteunen op het mentale
(cognitieve vlak) door informatie te aggregeren en te visualiseren. Alsook op
het fysieke vlak door taken over te nemen die te gevaarlijk of belastend zijn.
DECENTRALISATIE
• Toenemende autonomie van system op het vlak van het nemen van
beslissingen en het uitvoeren van taken

3. TOENEMENDE NOODZAAK VAN LEREN

4. EEN COGNITIEF PERSPECTIEF
COMMUNICATINGINFLUENCING
STORING &
RETRIEVING
SORTING
FILTERING
PATTERN
RECOGNITION
MEANING DECISON-
MAKINGINTELLIGENCE

5. INTELLIGENCEARTIFICIAL
AI IMITEERT COGNITIEVE PROCESSEN
INFLUENCING
STORING &
RETRIEVING
PATTERN
RECOGNITION
MEANING DECISON-
MAKING
COMMUNICATING
SORTING
FILTERING

6. DOMME MACHINES
ONZE ALLEDAAGSE ERVARING LEERT ONS DAT
COMPUTERS EIGENLIJK EERDER DOMME LINEAIRE
MACHINES ZIJN.
Photo by Simson Petrol on Unsplash

7. EEN PRAKTISCH VOORBEELD
Name 1 Name 2
Harold Lander
Lander Vervoort
Xi Hui Se
Nouri Mohammed
Selma Al-Nouri
Vervoort Patricia
Van de velde Jan
Person
Jan Van De Velde
Lander Vervoort
Selma AlNouri
Mohamed Nouri
Se Xi-Hui
Patricia Vervoort
Harold Lander
MERGE

8. NATURE TO THE RESCUE

9. ARTIFICIELE NEURALE NETWERKEN
Source: Mitchell, Machine Learning, p87. –
http://cse-wiki.unl.edu/wiki/index.php/Artificial_Neural_Networks, CC BY-SA 3.0,
https://commons.wikimedia.org/w/index.php?curid=22329629

10. NON-LINEAIRE COMPLEXE INFORMATIE

11. EEN ONGESUPERVISEERD SENTIMENT
NEURON

12. HOGERE COGNITIE: SEMANTISCHE
INFERENTIE
ORIGINAL RESEARCH
published: 17 January 2018
doi: 10.3389/fpsyg.2017.02335
Edited by:
Stefan Frank,
Radboud University Nijmegen,
Netherlands
Reviewed by:
Milena Rabovsky,
Humboldt-Universität zu Berlin,
Germany
Willem Zuidema,
University of Amsterdam, Netherlands
*Correspondence:
Using Neural Networks to Generate
Inferential Roles for Natural
Language
Peter Blouw* and Chris Eliasmith
Center for Theoretical Neuroscience, University of Waterloo, Waterloo, ON, Canada
Neural networks have long been used to study linguistic phenomena spanning the
domains of phonology, morphology, syntax, and semantics. Of these domains, semantics
is somewhat unique in that there is little clarity concerning what a model needs to be
able to do in order to provide an account of how the meanings of complex linguistic
expressions, such as sentences, are understood. We argue that one thing such models
need to be able to do is generate predictions about which further sentences are likely
to follow from a given sentence; these define the sentence’s “inferential role.” We then
show that it is possible to train a tree-structured neural network model to generate very
simple examples of such inferential roles using the recently released Stanford Natural
Language Inference (SNLI) dataset. On an empirical front, we evaluate the performance
of this model by reporting entailment prediction accuracies on a set of test sentences not
present in the training data. We also report the results of a simple study that compares
human plausibility ratings for both human-generated and model-generated entailments
for a random selection of sentences in this test set. On a more theoretical front, we argue
in favor of a revision to some common assumptions about semantics: understanding
a linguistic expression is not only a matter of mapping it onto a representation that
somehow constitutes its meaning; rather, understanding a linguistic expression is mainly
a matter of being able to draw certain inferences. Inference should accordingly be at the
core of any model of semantic cognition.
Blouw and Eliasmith Inferential Roles
FIGURE 2 | Generating entailments with paired encoder and decoder DT-RNNs. The decoder network computes a probability distribution over words at each node,
conditioned on the sentence representation produced by the encoder. The parameters of both the encoder and decoder are trained via backpropagation through
structure using error derivatives supplied at each node in the decoding tree. The encoder and decoder trees are dynamically generated for each pair of sentences in
the training data.
For instance, the sentence “A bird is in a pond” can be used to
generate the sentence “A little bird is outside in a small pond” by
using a decoding tree with nodes for two additional adjectives and
an additional adverb. If a predicted entailment is shorter than an
input sentence, then it tends to describe a more general situation.
For instance, the sentence “A little bird is outside in a small pond”
can be used to generate the sentence “A bird is outside” by using
a simple decoding tree with four nodes.
Second, these capacities for specification and generalization
suggest that the inferential transitions codified by the model
can be either inductive or deductive in nature. For example, the
a network of entailments for every sentence tha
from the model’s vocabulary.
Of course, nothing guarantees that these in
appropriate for all of the sentences in a given la
be rather miraculous if a simple model trained
thousand entailment pairs managed to always g
inferential transitions in novel scenarios. The
some degree of fit between the inferential role
model and the inferential roles that govern th
language. The goal of model development, th
improve this degree of fit.
FIGURE 3 | A model-generated inferential network around the sentence “Some kids are wrestling on an inflatable raft.” Each inferential transition
generating a predicted entailment after encoding the sentence at the beginning of each arrow. The entire network is generated starting with only t
the center of the diagram, which is drawn from the SNLI test set. Different decoding trees are used to generate the different entailments from the

13. VAN RESUME PARSING NAAR RESUME
ANALYSE
• 1998 – 2001: Project Manager at Fruity,
Inc. Minnesota. Used PMBOK™ to run a
transition project …
• 2001 – 2006: PMO at ElectricCars,
GmBH, Dusseldorf. Developed the office
from scratch, using Agile and Scrum, ….
• 2007 – today: Director of digital
technology at THD, Dusseldorf.
establishing shared understanding of tasks, and maintaining good communication throughout the project.
Also quality assurance and knowledge gaps are aspects that were mentioned by both interviewees and
freelancers as hurdles in an outsourced project, whereas the latter laid even more emphasis on finding
freelancers with appropriate knowledge and skills. Privacy and confidentiality of data, however, were
mostly a concern of the interviewees; not as much by the surveyed freelancers. Hence, although
outsourcing entails the hope to save resources in terms of time, money, and employees, outsourcing the
project could require additional effort in terms of high setup costs and loss of time through additional
communication, briefing, and performing quality assurance checks.
Skills… …Data
scientists
have
…Data scientists think
that freelancers should
have
…Freelancers have
Mean (1-5) Std. Dev.
Data understanding 45% 50% 4.45*** 0.614
Communication 55% 45% 4.23*** 0.763
Documentation and
writing
- 10% 4.08*** 0.868
Presentation 15% 10% 4.00*** 0.914
Visualization 10% 40% 3.96*** 0.892
Mathematics 90% 80% 3.91*** 0.86
Statistical skills 90% 80% 3.83*** 0.925
Programming 75% 65% 3.80*** 1.06
Database 55% 80% 3.40*** 0.963
Text Mining - 15% 3.34** 1.043
Machine Learning 30% 20% 3.28* 1.18
Domain Knowledge 55% 65% Was asked directly
Experience 35% 30%
Was inferred from years of
experience and self-reported
expertise on a 5-point scale
Legend: ***: 0.1%, **: 1%, *: 5% significance
Table 2. Freelancer skills required according to data scientists compared with existing skills on
freelance platforms
Furthermore, interviewees and surveyed freelancers stated preprocessing data as the most suitable task
to outsource to online freelancers. However, it is also the task that entails significant difficulties when
outsourced. Additionally, data scientists noted data preprocessing to be the most tedious task and the
one that they would like to outsource. Despite the various obstacles in outsourcing this step they would
be eager to see solutions that would allow to overcome these obstacles and therefore reduce their
workload. Problems identified during the interviews for the data cleaning process were: (1) possibly
confidential data, (2) the necessity of domain knowledge to understand the data, (3) the fact that data
cleaning entails many tacit, subjective assumptions, and (4) the necessity of putting significant trust into
freelancers when handing them data. Moreover, (5) data cleaning is a complex process, which requires a
lot of customer contact and has to be repeated iteratively. Hence, the (6) coordinating effort with the
freelancer is significant, as it has to be constantly maintained, and (7) specifications and results have to

14. THE WAY YOU CONFIGURE A SYSTEM DETERMINES THE
FUTURE OF THAT SYSTEM.
AND SOMETIMES, YOU CAN SET UP THAT SYSTEM SO IT
DOES THINGS
HUMANS LIKE.
Uit: Soonish - Ten emerging technologies that’ll improve and/or ruin everything
Kelly & Zach Weinersmith

15. MACHINE
MAN
MACHINE
MACHINE
MAN
MACHINE
MAN
MAN
HR LOCUS OF CONTROL
OPERATIONS:LOCUSOFCONTROL
HR TODAY
HR
TOMORROW
HR
TOMORROW