ANN3(CNN + DL) Flashcards

1
Q

Deep learning ANN:s har många lager. Men en ANN kan ha många lager utan att vara en DL. Vad är det som utmärker en DL ANN från andra ANN:S?

A

it is typically viewed as (hierarchically) decomposing the problem
into features

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Varför ses det som ett problem att det är svårt att förstå varför ett DL network får fram den
outputen den får fram?

A

For workers using AI in safety critical job areas, e.g. healthcare, self-driving car development, it is important to know why the DNN makes its
decision (e.g. in case it makes a mistake).

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Nämn några “emerging technologies” som DL ligger bakom

A

Autonomous Vehicles, Face and Emotion recognition software, Online object recognition and description
Det används också för att förbättra sånt som Gaming opponents, natural language processing(GPT) och robotics

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

När skedde “the deep learning revolution”?

A

2012

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Vad finns det för anledningar till att deep learning inte blev grej tidigare?

A

In the 90s and 00s we had problems of:
*Computational Power
*Availability of Data sets
*Inefficient learning algorithms

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What has increased the computational power since the 90s/00s and enabled the DL revolution?

A
  • CPU:s power has been increasing since the 70’s, but in the 00s Nvidia GPU:s(graphic processing units) came that speeds up processing of “big data”
  • In 2016 came TPUs (tensor processing units) – designed for fast but low precision
    computation (specialized for ANN computation)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Vad har löst problemen med tillgänglighet till data(för DL-träning)?

A
  • Availability of big (free) datasets on the internet
  • Algorithm sharing (e.g. via Github) people develop their learning algorithms
  • Benchmarking new algorithms: new algorithms can be tested on the available big datasets
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Vad upptäckte Hubel och Wiesel 1959?

A

Found single cells in primary visual cortex) sensitive to lines at different orientations. important for visual related AI

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

När kom den första CNN och vad hette den?

A

LeNet-5 architecture developed by LeCun et al. (1998)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Vad var det för nätverk som Krizhevsky, Sutskever & Hinton tog fram 2012?

A

AlexNet: A type of Deep Convolutional Neural Network

How well did you know this?
1
Not at all
2
3
4
5
Perfectly