ai for document understanding - data innovation lab: tum ... · dataset • rvl-cdip dataset letter...
TRANSCRIPT
![Page 1: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/1.jpg)
Final Presentation – 18.02.2019
AI for Document Understanding
![Page 2: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/2.jpg)
TUM-DI-LAB Team
Wassim OliverCingisAlican
Studies:Master Mathematics in Data
Science
Studies:Master Mathematics
Studies:Master Mathematical Finance
and Actuarial Science
Studies:Master Mathematics
2
![Page 3: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/3.jpg)
Hey, my name is John.
3
![Page 4: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/4.jpg)
I work at a big accounting
consulting company.
4
![Page 5: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/5.jpg)
Today is a bad day.
5
![Page 6: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/6.jpg)
Investigate an audit of a big firm.
6
![Page 7: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/7.jpg)
Done in 1995. No digital trace.
7
![Page 8: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/8.jpg)
Letters.
8
![Page 9: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/9.jpg)
Receipts.
9
![Page 10: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/10.jpg)
Invoices.
10
![Page 11: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/11.jpg)
Emails.
11
![Page 12: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/12.jpg)
Thousands of them.
12
![Page 13: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/13.jpg)
How it feels like?
13
![Page 14: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/14.jpg)
14[17]
![Page 15: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/15.jpg)
[1],[2],[3]
15
![Page 16: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/16.jpg)
[1],[2],[3][1],[2],[3]
16
![Page 17: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/17.jpg)
EmailEmailEmail
Task
EmailLetter
EmailLetter
17
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
![Page 18: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/18.jpg)
Dataset
• RVL-CDIP dataset
EmailLetter
[14]
[15]
Memo Filefolder Form Handwritten Invoice Advertisement
Budget News Article
Presentation ScientificPublication
Questionnaire Resume ScientificReport
Specification
18
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
![Page 19: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/19.jpg)
First Steps
Graph-Based [11] Content-Based[12]
Image-Based [13]
19
+ Very good accuracy
- Complicated preprocessing
+ State-of-the-art+ Simple preprocessing+ Matches our project goals
- Training time
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
+ Unconventional approach
- Complicated preprocessing- Hard to implement
![Page 20: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/20.jpg)
Classification Model
20
• Validation accuracy: 0.98• Test accuracy: 0.98
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
![Page 21: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/21.jpg)
Preprocessing
21
LeftHeader Holistic RightBottom
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
![Page 22: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/22.jpg)
Region-Based & Holistic CNNs
22
Transfer Learning
Classification Model Task Dataset First Steps Classification Model PreprocessingRegion-Based &
Holistic CNNs
![Page 23: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/23.jpg)
[1],[2],[3]
23
![Page 24: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/24.jpg)
OCR Illustration
OCR Illustration Pipeline Preprocessing & Postprocessing Usefulness
24
![Page 25: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/25.jpg)
OCR Illustration
OFFICES IN: NEW YORK.DETROIT.LOS ANGELES.WASNINGTON D C.CHICAGO AND OTHER PRINCIPAL CITIES While Padro TV Reoriis imc endeevors to ...
PROGRAM NEWS 60 secs. STATION CBS DATE JUNE 11, 1968 7:21 PM CITY NEW YORK
KENT CIGARETTES 803231(srx TRAIN WHISTLE)(MUSIC)MAN: Here’s the story they still tell around the railroad yard, about one nlqht when Casey Jones was drlvln’ extra hard. Another train ahead, but watch out. An accident! But Casey wouldn’t jump
OCR Illustration Pipeline Preprocessing & Postprocessing Usefulness
25
![Page 26: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/26.jpg)
Accuracy Metric
Image Preprocessing
OCR Pipeline
26
OCR Illustration Pipeline Preprocessing & Postprocessing Usefulness
DocumentImage
PyTesseractOutput as
StringPostprocessing
![Page 27: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/27.jpg)
Preprocessing & Postprocessing
Preprocessing
• Upscaling
• Sharpness
• Contrast
Postprocessing
• Autocorrection
• Stopwords
27
OCR Illustration Pipeline Preprocessing & Postprocessing Usefulness
![Page 28: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/28.jpg)
Factors trained on
Usefulness ofPreprocessing
28
OCR Illustration Pipeline Preprocessing & Postprocessing Usefulness
![Page 29: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/29.jpg)
[1],[2],[3]
29
![Page 30: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/30.jpg)
Research – NER Algorithms
•No handcrafted features
•State-of-the-art results
•Mathematically interesting
•Curiosity about new networks
Decision for NNs
•4 different papers
•Networks are comparable, yet different
Read papers•Embeddings and CNN
•BiLSTM
•CRF
•Output of NER (.json)
Implementation
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
30
![Page 31: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/31.jpg)
Named Entities: Organization Location Person
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
Named Entity Recognition
31
![Page 32: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/32.jpg)
Facebook Inc. is endowing a new institute, led by
Christoph Luetge , devoted to the ethics of artificial
intelligence at the Technical University of Munich ,
in Germany .
Named Entities: Location Person
[10]
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
Named Entity Recognition
32
Organization
![Page 33: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/33.jpg)
Facebook Inc. is endowing a new institute, led by
Christoph Luetge , devoted to the ethics of artificial
intelligence at the Technical University of Munich ,
in Germany .
Location PersonNamed Entities:
Named Entity Recognition
[10]
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
33
Organization
![Page 34: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/34.jpg)
BiLSTM-CNN-CRF: Complete Architecture
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
CasingInformation
BiLSTM CRFSequence of Labels for the Sentence
Word Embeddings
Sentence
CharacterRepresentation
34
• Validation accuracy: 94.2%• Test accuracy: 90.8%
![Page 35: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/35.jpg)
CNN: Transform character-level information into character-representation
CNN-Extracted Features
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
35
![Page 36: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/36.jpg)
GloVe: Projection into 2D
Word Embeddings:• Pretrained word embeddings with GloVe
100 dimensions.
Word-Embeddings using GloVe
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
[16]
36
![Page 37: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/37.jpg)
The casing of a word contains important information
1994
mainly numericnumeric all lower all upper
N26 TUMhouse Munich
initial upper
Casing Information
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
37
Casing:
Embedding
![Page 38: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/38.jpg)
Bi-Directional Long Short-Term Memory
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
Capture context on both sides of a word of a sentence
Forward LSTM
Backward LSTM
Artificial intelligence for document understanding
Output
38
CharacterRepresentation
Word Embedding
CasingInformation
![Page 39: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/39.jpg)
[…] Technical University of Munich, in Germany.
Which sequence of labels is most likely for the sentence?
Conditional Random Field
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
39
![Page 40: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/40.jpg)
[…] Technical University of Munich, in Germany.
Which sequence of labels is most likely for the sentence?
Not likely:
Conditional Random Field
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
40
[…] Technical University of Munich, in Germany .Organization Person LocationLocation
![Page 41: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/41.jpg)
[…] Technical University of Munich, in Germany.
Which sequence of labels is most likely for the sentence?
Not likely:
More likely:
✔ […] Technical University of Munich, in Germany .Organization Location
Conditional Random Field
NER ResearchNamed Entity Recognition
CompleteArchitecture
CharacterRepresentation
GloVeCasing
InformationBiLSTM
ConditionalRandom Field
41
[…] Technical University of Munich, in Germany .Organization Person LocationLocation
![Page 42: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/42.jpg)
[1],[2],[3]
42
![Page 43: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/43.jpg)
BananaDashboard
43
Search Module
[19]
[20]
[3]
[21]
[22]
[23], [24]
![Page 44: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/44.jpg)
Classification [4] Optical CharacterRecognition [5]
Named Entity Recognition [6]
Search Module
OpenNebula [7] Flask [8] Continuous Integration – Jenkins [9]
Goals and Outcomes
44
![Page 45: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/45.jpg)
Open Nebula
Controller
Filesystem
Image Classification
Optical Character
Recognition (OCR)
NamedEntity
Recognition (NER)
Solr & Banana
Cloud Structure (Flask)
45
![Page 46: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/46.jpg)
Thank you for your Attention
46
![Page 47: AI for Document Understanding - Data Innovation Lab: TUM ... · Dataset • RVL-CDIP dataset Letter Email [14] [15] Memo Filefolder Form Handwritten Invoice Advertisement Budget News](https://reader030.vdocument.in/reader030/viewer/2022041014/5ec582f09f76806f70487aa7/html5/thumbnails/47.jpg)
References Logo OpenNebula, https://opennebula.org/referencing/, 2019-02-08 [1] Logo Flask, https://de.wikipedia.org/wiki/Datei:Flask_logo.svg, 2019-02-08 [2] Logo Solr, http://lucene.apache.org/solr/logos-and-assets.html, 2019-02-08 [3] Classification by cre.ativo mustard from the Noun Project [4] Content by Jyoti Vyas from the Noun Project [5] Logo NER, https://wordlift.io/blog/en/entity/named-entity-recognition/, 2019-02-13 [6] Logo OpenNebula, https://www.v3.co.uk/v3-uk/news/2356529/microsoft-gets-more-open-with-cross-platform-packerio-and-
opennebula-cloud-tools, 2019-02-13 [7] Logo Flask, https://engineering.bitnami.com/articles/deploy-a-production-ready-mariadb-cluster-on-kubernetes-with-bitnami-and-
helm.html, 2019-02-14 [8] process by Gregor Cresnar from the Noun Project [9] Facebook Endows AI Ethics Institute at German University TUM, Jeremy Kahn, https://www.bloomberg.com, 2019-01-20 [10] Graph, https://thenounproject.com/mb.icons/collection/network/?i=1775952, 2019-02-14 [11] SVM-SVD, https://thenounproject.com/search/?q=svm&i=1503831, 2019-02-14 [12] Network, https://thenounproject.com/smodgekar/collection/technology/?i=1714861, 2019-02-14 [13] RVL-CDIP sample1, http://www.cs.cmu.edu/~aharley/rvl-cdip/images/sample1.png, 2019-02-16 [14] RVL-CDIP sample2, http://www.cs.cmu.edu/~aharley/rvl-cdip/images/sample2.png, 2019-02-16 [15] GloVe, https://nlp.stanford.edu/projects/glove/, 2019-02-16 [16] Paperwork, https://de.depositphotos.com/62875115/stock-illustration-vector-paperwork-mood.html, 2019-02-17 [17] Banana, https://www.alamy.com/banana-logo-template-vector-icon-illustration-design-image159202591.html, 2019-02-17 [19] Dashboard, https://doc.lucidworks.com/fusion/2.0/Dashboards.html, 2019-02-17 [20] Magnifier, Search by Mello from the Noun Project, 2019-02-17 [21] Data, database by Aiden Icons from the Noun Project, 2019-02-17 [22] Cloud-Structure, cloud storage by un·delivered from the Noun Project, 2019-02-17 [23] Cloud, implementation by Tomas Knopp from the Noun Project, 20189-02-17 [24]
47