SlideShare a Scribd company logo
1 of 22
Presentation at Global AI Conference
Santa Clara Convention Center
Santa Clara, CA
January 23, 2020
Jeff Shomaker
Founder, 21 SP, Inc.
Natural Language
Processing (NLP):
Future Trends
21 SP, Inc.
Proprietary and Confidential 2
Introduction
• Natural Language Processing (NLP) is making rapid progress
• Talk will cover:
– A little history
– What is NLP and what can you do with it?
– Popular NLP Models
– Major Model Advancements in 2019
– Top 10 AI Use Cases by Revenue
– Commercial Products Announced in 2019
– Future Trends for NLP
21 SP, Inc.
Proprietary and Confidential
How New is NLP?
– In 1637, Descartes wrote that language model-based
machines would likely be possible. He said:
For one may conceive that a machine would be so made
that it could generate speech…but not so that it could adapt
its output in a sufficiently versatile way so as to answer the
meaning of everything that could be said in its presence, as
even the stupidest of man can. 1)
(1) Francois Chollet (2019 Nov 19). @fchollet, Tweet, See Bibliography for full citation.
3
21 SP, Inc.
Proprietary and Confidential
What Can NLP Do?
– NLP can be used for the following:
• Question answering
• Speech recognition
• Text-to-speech and speech-to-text
• Topic modeling
• Sentiment classification
• Language modeling
• Translation
• Others. 2)
(2) Rachael Thomas (2019 May-Jun). fast.ai Code-First Intro to Natural Language
Processing, youtube, accessed 11-11-19. See bibliography.
4
21 SP, Inc.
Proprietary and Confidential
NLP Models
– In the past, machine learning-based NLP processing was done with
support vector machines (SVM) and logistic regression.
– Recently, neural network (NN) models have been increasingly used
and have provided better results on some NLP tasks.
– Deep learning NNs use a “… multi-level automatic feature
representation….” approach.
– In the following, we will discuss four types of models used for NLP:
• Convolutional neural networks (CNNs)
• Recurrent neural networks (RNNs)
• Recursive neural networks (Recur NNs) 3)
• Transformer models. 17)
(3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography.
(17) Ashish Vaswani, et al (2017 Dec 6). Attention is all you need [On Transformers]. See
bibliography.
5
21 SP, Inc.
Proprietary and Confidential
NLP: Convolutional Neural Networks (CNNs)
– Collobert and Weston proposed CNNs for sentence modeling and
other NLP tasks in 2008 and 2011 respectively.
– Word embeddings - an approach that identifies which words are
similar and the relationships between them.
– After word embeddings came about, it was necessary to extract “…
higher-level features from constituting words or n-grams.” 20)
– NLP models use these for translation, question answering,
summarization and sentiment analysis.
– CNNs are good for capturing semantic information.
– CNNs are not so good when data scarcity exists and they struggle
when modeling some contextual information. 3)
(3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See Bibliography.
(20) N-grams are a sequence of words, eg, a 2-gram might be “please turn”. S. Kapadia (2019 Mar 26).
Language Models: N-Gram, See bibliography. .
6
21 SP, Inc.
Proprietary and Confidential
NLP: Recurrent Neural Networks (RNN)
– RNN design (ie, sequential information processing) makes them good for
NLP since they do the same thing on each instance of the sequence. 3)
– Since RNNs process in a looping fashion and they reuse weights,
information is remembered during processing. 4)
– The process works well for language modeling, machine translation,
speech recognition and image capturing.
– It is not clear whether RNNs or CNNs are best for sentiment
classification, question answering or part-of-speech-tagging.
– One drawback with simple RNNs is they struggle in learning and tuning
an NN’s early layers. LSTMs (ie, RNN variant) overcome this issue. 3)
(3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography.
(4) Vincent Boucher (2019 Nov 25). AI Overview, academy.montreal.ai. See bibliography.
7
21 SP, Inc.
Proprietary and Confidential
NLP: Recursive Neural Networks (Recur NNs)
– Recursive NNs use a tree-structure to account for the fact that in
language “… words and sub-phrases combine into phrases in a
hierarchical manner.”
– Recur NNs work well for parsing* and there are two types:
• Recursive autoencoders – can reconstruct context
• Recursive neural tensor network – at each node in a tree, they
compute a supervised objective. 3)
– Recur NNs can be used for the following applications:
• NLP
• Audio-to-text transcription
• Image scene decomposition. 5)
* Parsing – “Analyze (a string or text) into logical syntactic components, typically in order to test conformability to a
logical grammar.” LEXICO, lexico.com.
(3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography.
(5) J. Patterson & A. Gibson (2017). Deep Learning: A Practitioner’s Approach. See bibliography.
8
21 SP, Inc.
Proprietary and Confidential
NLP: Transformers
– The Transformer model is a new approach to processing text that
changed thinking and enabled Google’s BERT, which set the standard.
16), 18)
– The Transformer uses only attention mechanisms, which allow them to
avoid recurrence and convolutions.
– This new architecture makes it possible to process in a parallel fashion,
which can speed up training significantly over RNNs.
– In this paper, which proposed this innovative architecture, Transformers
were tested on translation tasks using BLEU:
• For English-to-German, results showed improvement over all other
models
• For English-to-French, results bested all earlier single models. 17)
(16) Prateek Joshi (2019 Jun 19). How do Transformers Work in NLP? See bibliography.
(17) Ashish Vaswani, et al (2017 Dec 6). Attention is all you need [On Transformers]. See bibliography.
(18) Jacob Devlin, et al (2019 May 24 – 1st version 2018 Oct 11). BERT: Pre-training of Deep Bidirectional
Transformers. See bibliography. On 11 NLP tasks, BERT beat all earlier models.
9
21 SP, Inc.
Proprietary and Confidential
2019 Advancement: XLNET
– XLNET is a “... Generalized autoregressive pretraining method ….” that
allows learning in a bidirectional manner and has improved performance
compared to BERT.
– XLNET has achieved excellent results on:
• Language understanding
• Reading comprehension
• Test classification
• Document ranking.
– Using many benchmarks, XLNET is superior to BERT. 11)
– Potential business applications include:
• Automated customer support
• Sentiment analysis (eg, brand awareness)
• Information searching in document data bases. 12)
(11) Zhilin Yang, et al (2019 Jun 19). XLNET: Generalized Autoregressive Pretraining. See bibliography.
(12) Mariya Yao (2019 Nov 12). Major NLP Achievements in 2019. See bibliography.
10
21 SP, Inc.
Proprietary and Confidential
2019 Advancement: ALBERT – A Lite BERT
– ALBERT is an architecture that uses a transformer encoder.*
– ALBERT scales better than BERT due to parameter reductions and an
improved method for predicting sentences.
– With fewer parameters, memory needs are reduced and training speed is
increased.
– ALBERT delivers state-of-the-art performances on GLUE, RACE, and
SQuAD tests. 15)
– Potential business applications include:
• Chatbots
• Sentiment analysis
• Document mining
• Text classification. 12)
*Transformers are a major improvement over RNN-based sequence-to-sequence models used previously.
(15) Zhenzhong Lan, et al (2019 Oct 30). ALBERT: A Lite BERT for Self-Supervised Learning. See bibliography.
(12) Mariya Yao (2019 Nov 12). Major NLP Achievements in 2019. See bibliography.
11
21 SP, Inc.
Proprietary and Confidential
2019 Advancement: DistilBERT – For the Edge
– Transfer learning is used more and more in NLP; but, these big models
are hard to train to do inference on-the-edge (eg, small devices). 21)
– These large models can have parameters in the hundreds of millions.
– Paper’s authors propose an approach for a “… general purpose language
representation model .…” that can run on-the-edge.
– DistilBERT use knowledge distillation which “… is a compression
technique in which a compact model – the student – is trained to
reproduce the behavior of a larger model – the teacher – or an ensemble
of models.”
– They discuss models that are:
• 40% smaller than BERT
• Keep 97% of BERT’s language understanding
• Are 60% faster than BERT. 13)
(13) Victor Sanh, et al (2019 Oct 16). DistilBERT, A distilled version of BERT. See bibliography.
(21) Dipanjan (DJ) Sarkar (2018 Nov 14). A Comprehensive Hands-on Guide to Transfer Learning. See
bibliography. Transfer learning allows information gained from an earlier model to be used for training a
subsequent model (eg, reuse weights).
12
21 SP, Inc.
Proprietary and Confidential
AI Revenue -Top Ten Use Cases Forecast 6)
13
(6) Mike Quindazzi (2019 Dec 7). Cumulative AI Software Revenue, @tractica via @mikequindazzi,
Tweet, See bibliography.
21 SP, Inc.
Proprietary and Confidential
2019 New Product: Qualcomm’s Chipset
– High end Android phones will soon have a new chipset called the
Snapdragon 865.
– The chip will make possible 5G, digital drivers licenses, a 200 megapixel
camera and improved gaming on smartphones.
– A very notable feature is “Instant live translation and transcription.”
– Google Translate already has very good quality -- Android 10 provides
live video transcriptions and Pixel 4 does translations with it’s voice
recorder app.
– It is reported that the demonstrations with the 865 are far superior. They
show:
• On-the-fly translation in English and Chinese
• Words spoken in English can be translated and then spoken in
Chinese to the listener. 9)
(9) Jessica Dolcourt (2019 Dec 19). 6 Things Coming to Android Phones, CNET. See bibliography.
14
21 SP, Inc.
Proprietary and Confidential
2019 New Product: Automatic Transcription
– Amazon has brought out a product called Transcribe Medical
• Uses a voice app and Amazon’s Web Services to automatically
transcribe patient-doctor discussions
• Once the discussion is complete, a transcription is placed in patient’s
records.
– Patient health records can be further analyzed by Amazon’s Comprehend
Medical product
• This analyzes text for patterns that can be used to highlight possible
diagnoses and approaches to treatment.
– Nuance and Microsoft collaborated earlier and deployed an improved
Dragon Medical Virtual Assistant using Microsoft’s Azure platform.
– Google is also developing a medical transcription product with Stanford
University. 8)
(8) Eric H. Schwartz (2019 Dec 6). Amazon Launches Medical Transcription, voicebot.ai. See bibliography.
15
21 SP, Inc.
Proprietary and Confidential
2019 New Product: AI-enabled Robot w/ Emotion
– CIMON (Crew Interactive Mobile Companion) was the first smart
astronaut assistant used on the International Space Station (ISS).
– CIMON is a collaboration between IBM (Watson AI), Airbus, and the
German Aerospace Center.
– The robotic assistant is designed to retrieve information, track tasks, and
with the new version, CIMON-2, mitigate issues between crew members.
– CIMON-2, which was sent aboard December 2019’s SpaceX ISS
resupply mission, now uses IBM Watson’s Tone Analyzer.
– The goal is to analyze astronaut’s conversations to allow for:
• The mitigation of group-think
• Provide an objective perspective or contrarian viewpoint when
complicated decisions need to be made. 10)
(10) D. Etherington (2019 Dec 5). AI-enabled Assistant Returns to Space, Techcrunch. See bibliography.
16
21 SP, Inc.
Proprietary and Confidential
A 2020 NLP Wish List
– Sebastion Ruder’s list includes:
• Learning from few samples rather than large datasets
• Compact and efficient rather than huge models
• Evaluate on at least another language (from a different language
family)
• New datasets contain at least one other language
• NLP helps to unlock scientific knowledge (see Nature 19)). 7)
(7) Sebastian Ruder (2019 Dec 23). 2020 Wish Lists, NLP News, Issue #48. See bibliography.
(19) Vahe Tshitoyan, et al (2019 Jul 3). Unsupervised learning captures latent knowledge. See bibliography.
17
21 SP, Inc.
Proprietary and Confidential
2020 AI and Machine Learning Trends
– Natural Language Processing
• Pre-training and fine tuning is the approach
• Improvements in NLP are likely to be driven by the use of linguistics
and the incorporation of more human knowledge
• Neural machine translation (ie, simultaneous machine translation) is
now practical and is expected to improve further
– Conversational AI
• Chatbots are widely used for customer service
• To improve Chatbots, researchers are:
– Leveraging conversational history and context
– Working towards providing more varied responses
– Attempting to incorporate emotion and empathy into
conversations. 14)
(14) Kate Koidan (2020 Jan 2). Top AI Trends for 2020, TopBots. See bibliography.
18
21 SP, Inc.
Proprietary and Confidential
End Notes
• 1) Francois Chollet [Google Deep Learning] (2019 Nov 19), @fenollet, Tweet, accessed 12-18-19.
• 2) Rachael Thomas (2019 May-Jun). fast.ai Code-First Intro to Natural Language Processing, What is
NLP? (NLP Video 1), youtube, accessed 11-11-19.
• 3) Tom Young, et al (2018 Nov 25). Recent Trends in Deep Learning Based Natural Language
Processing, ar.Xiv:1708.02709v8 [cs.CL], accessed November 2019.
• 4) Vincent Boucher (2019 Nov 25). Montreal.AI Academy: Artificial Intelligence 101 First World-Class
Overview of AI for All, Montreal.AI, academy.montreal.ai, accessed 12-28-19.
• 5) Josh Patterson and Adam Gibson, Deep Learning: A Practitioner’s Approach, O’Reilly 2017,
accessed 12-22-19 at allite.books.com.
• 6) Mike Quindazzi (2019 Dec 7). Cumulative Artificial Intelligence Revenue, Top Ten Use Cases
World Markets:2017-2025, @tractica via @mikequindazzi, Tweet, accessed 12-29-19.
• 7) Sebastian Ruder [Deep Mind AI] (2019 Dec 23). 2020 Wish Lists, Natural Language Processing
News, Issue #48, newsletter.ruder.io, accessed 12-31-19.
• 8) Eric Hal Schwartz (2019 Dec 6). Amazon Launches Medical Transcription Service in Direct
Competition with Nuance [Nuance partner is Microsoft], voicebot.ai, accessed 12-30-19.
• 9) Jessica Dolcourt (2019 Dec 12). 6 Things Coming to 2020 Android Phones: 8K Video, 5G Uploads,
Two-finger Unlock, CNET, cnet.com, accessed 1-1-20.
19
21 SP, Inc.
Proprietary and Confidential
End Notes (cont.)
• 10) Darrell Etherington (2019 Dec 5). AI-enabled Assistant Robot Returning to the Space Station with
Improved Emotional Intelligence, Techcrunch, techcrunch.com, accessed 12-31-19.
• 11) Zhilin Yang, et al (2019 Jun 19). XLNET: Generalized Autoregressive Pretraining for Language
Understanding. arXiv:1906.08237v1 [cs.CL].
• 12) Mariya Yao (2019 Nov 12). What Are Major NLP Achievements & Papers from 2019, TopBots,
topbots.com, accessed 12-31-19.
• 13) Victor Sanh, et al (2019 Oct 16). DistilBERT, a Distilled Version of BERT: Smaller, Faster,
Cheaper and Lighter, arXiv:1910.01108v2 [cs.CL], accessed 1-8-20.
• 14) Kate Koidan (2020 Jan 2). Top AI & ML Research Trends for 2020, TopBots, topbots.com,
accessed 1-5-20.
• 15) Zhenzhong Lan, et al (2019 Oct 30). ALBERT: A Lite BERT for Self-Supervised Learning of
Language Representations, arXiv:1909.11942v3 [cs.CL], accessed 1-2-20.
• 16) Prateek Joshi (2019 Jun 19). How do Transformers Work in NLP: A Guide to the Latest State-of-
the-Art Models, Analytics Vidhya, analyticsvidhya.com, accessed 1-10-20.
• 17) Ashish Vaswani, et al (2017 Dec 6). Attention Is All You Need [Proposes The Transformer model],
arXiv:1706.03762v5 [cs.CL].
• 18) Jacob Devlin, et al (2019 May 24 – 1st version 2018 Oct 11). BERT: Pre-training of Deep
Bidirectional Transformers for Language Understanding, arXiv:1810.04805v2 [cs.CL].
20
21 SP, Inc.
Proprietary and Confidential
End Notes (cont.)
• 19) Vahe Tshitoyan, et al (2019 Jul 3). Unsupervised Word Embeddings Capture Latent Knowledge
from Materials Science Literature, Nature, nature.com, accessed 1-17-20.
• 20) Shashank Kapadia (2019 Mar 26). Language Models: N-Gram, Towards Data Science,
towardsdatascience.com, accessed 1-2-20.
• 21) Dipanjan (DJ) Sarkar (2018 Nov 14). A Comprehensive Hands-on Guide to Transfer Learning with
Real-World Applications in Deep Learning, Towards Data Science, towardsdatascience.com,
accessed 1-19-20.
•
21
21 SP, Inc.
Proprietary and Confidential
Contacts
• Jeff Shomaker – Founder/President 21 SP, Inc.
– jshomaker@21spinc.com
– www.21spinc.com
– 650-285-8122
• 21 SP, Inc. is a small privately held startup developing and marketing
expert systems-based decision support software to use in genetic-
based personalized medicine. The company's mission is to create
tools that will reduce the use of traditional trial-and-error medicine by
using pharmacogenetics and other evidence-based data, such as the
results of high quality clinical trials, in the medical clinic.
22

More Related Content

What's hot

Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
Grammarly
 
NLP_Project_Paper_up276_vec241
NLP_Project_Paper_up276_vec241NLP_Project_Paper_up276_vec241
NLP_Project_Paper_up276_vec241
Urjit Patel
 
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATIONTHE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
ijscai
 
CSTalks-Natural Language Processing-17Aug
CSTalks-Natural Language Processing-17AugCSTalks-Natural Language Processing-17Aug
CSTalks-Natural Language Processing-17Aug
cstalks
 

What's hot (18)

[Paper Reading] Supervised Learning of Universal Sentence Representations fro...
[Paper Reading] Supervised Learning of Universal Sentence Representations fro...[Paper Reading] Supervised Learning of Universal Sentence Representations fro...
[Paper Reading] Supervised Learning of Universal Sentence Representations fro...
 
A prior case study of natural language processing on different domain
A prior case study of natural language processing  on different domain A prior case study of natural language processing  on different domain
A prior case study of natural language processing on different domain
 
Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
Grammarly AI-NLP Club #2 - Recent advances in applied chatbot technology - Jo...
 
NLP_Project_Paper_up276_vec241
NLP_Project_Paper_up276_vec241NLP_Project_Paper_up276_vec241
NLP_Project_Paper_up276_vec241
 
Intelligence and artificial intelligence
Intelligence and artificial intelligenceIntelligence and artificial intelligence
Intelligence and artificial intelligence
 
Conversational AI:An Overview of Techniques, Applications & Future Scope - Ph...
Conversational AI:An Overview of Techniques, Applications & Future Scope - Ph...Conversational AI:An Overview of Techniques, Applications & Future Scope - Ph...
Conversational AI:An Overview of Techniques, Applications & Future Scope - Ph...
 
Artificial intelligence and knowledge representation
Artificial intelligence and knowledge representationArtificial intelligence and knowledge representation
Artificial intelligence and knowledge representation
 
Introduction to AI
Introduction to AIIntroduction to AI
Introduction to AI
 
EXTENDING OUTPUT ATTENTIONS IN RECURRENT NEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENT NEURAL NETWORKS FOR DIALOG GENERATIONEXTENDING OUTPUT ATTENTIONS IN RECURRENT NEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENT NEURAL NETWORKS FOR DIALOG GENERATION
 
COLING 2014: Joint Opinion Relation Detection Using One-Class Deep Neural Net...
COLING 2014: Joint Opinion Relation Detection Using One-Class Deep Neural Net...COLING 2014: Joint Opinion Relation Detection Using One-Class Deep Neural Net...
COLING 2014: Joint Opinion Relation Detection Using One-Class Deep Neural Net...
 
Code Mixing computationally bahut challenging hai
Code Mixing computationally bahut challenging haiCode Mixing computationally bahut challenging hai
Code Mixing computationally bahut challenging hai
 
French machine reading for question answering
French machine reading for question answeringFrench machine reading for question answering
French machine reading for question answering
 
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATIONTHE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
THE EFFECTS OF THE LDA TOPIC MODEL ON SENTIMENT CLASSIFICATION
 
AI_ 3 & 4 Knowledge Representation issues
AI_ 3 & 4 Knowledge Representation issuesAI_ 3 & 4 Knowledge Representation issues
AI_ 3 & 4 Knowledge Representation issues
 
NLP Bootcamp 2018 : Representation Learning of text for NLP
NLP Bootcamp 2018 : Representation Learning of text for NLPNLP Bootcamp 2018 : Representation Learning of text for NLP
NLP Bootcamp 2018 : Representation Learning of text for NLP
 
Knowledge representation
Knowledge representationKnowledge representation
Knowledge representation
 
Warnikchow - SAIT - 0529
Warnikchow - SAIT - 0529Warnikchow - SAIT - 0529
Warnikchow - SAIT - 0529
 
CSTalks-Natural Language Processing-17Aug
CSTalks-Natural Language Processing-17AugCSTalks-Natural Language Processing-17Aug
CSTalks-Natural Language Processing-17Aug
 

Similar to Nlp 2020 global ai conf -jeff_shomaker_final

Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdfTransfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
oranisalcani
 
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
Sharmila Sathish
 
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
AishwaryaChemate
 
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
kevig
 

Similar to Nlp 2020 global ai conf -jeff_shomaker_final (20)

Natural Language Processing - Research and Application Trends
Natural Language Processing - Research and Application TrendsNatural Language Processing - Research and Application Trends
Natural Language Processing - Research and Application Trends
 
Working in NLP in the Age of Large Language Models
Working in NLP in the Age of Large Language ModelsWorking in NLP in the Age of Large Language Models
Working in NLP in the Age of Large Language Models
 
SENTIMENT ANALYSIS IN MYANMAR LANGUAGE USING CONVOLUTIONAL LSTM NEURAL NETWORK
SENTIMENT ANALYSIS IN MYANMAR LANGUAGE USING CONVOLUTIONAL LSTM NEURAL NETWORKSENTIMENT ANALYSIS IN MYANMAR LANGUAGE USING CONVOLUTIONAL LSTM NEURAL NETWORK
SENTIMENT ANALYSIS IN MYANMAR LANGUAGE USING CONVOLUTIONAL LSTM NEURAL NETWORK
 
Sentiment Analysis In Myanmar Language Using Convolutional Lstm Neural Network
Sentiment Analysis In Myanmar Language Using Convolutional Lstm Neural NetworkSentiment Analysis In Myanmar Language Using Convolutional Lstm Neural Network
Sentiment Analysis In Myanmar Language Using Convolutional Lstm Neural Network
 
Thomas Wolf "An Introduction to Transfer Learning and Hugging Face"
Thomas Wolf "An Introduction to Transfer Learning and Hugging Face"Thomas Wolf "An Introduction to Transfer Learning and Hugging Face"
Thomas Wolf "An Introduction to Transfer Learning and Hugging Face"
 
A-STUDY-ON-SENTIMENT-POLARITY.pdf
A-STUDY-ON-SENTIMENT-POLARITY.pdfA-STUDY-ON-SENTIMENT-POLARITY.pdf
A-STUDY-ON-SENTIMENT-POLARITY.pdf
 
Analysis of the evolution of advanced transformer-based language models: Expe...
Analysis of the evolution of advanced transformer-based language models: Expe...Analysis of the evolution of advanced transformer-based language models: Expe...
Analysis of the evolution of advanced transformer-based language models: Expe...
 
ijeter35852020.pdf
ijeter35852020.pdfijeter35852020.pdf
ijeter35852020.pdf
 
Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdfTransfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
Transfer_Learning_for_Natural_Language_P_v3_MEAP.pdf
 
State of the art in Natural Language Processing (March 2019)
State of the art in Natural Language Processing (March 2019)State of the art in Natural Language Processing (March 2019)
State of the art in Natural Language Processing (March 2019)
 
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
Feature Extraction and Analysis of Natural Language Processing for Deep Learn...
 
The Smart Way To Invest in AI and ML_SFStartupDay
The Smart Way To Invest in AI and ML_SFStartupDayThe Smart Way To Invest in AI and ML_SFStartupDay
The Smart Way To Invest in AI and ML_SFStartupDay
 
Neel Sundaresan - Teaching a machine to code
Neel Sundaresan - Teaching a machine to codeNeel Sundaresan - Teaching a machine to code
Neel Sundaresan - Teaching a machine to code
 
IRJET - Automatic Lip Reading: Classification of Words and Phrases using Conv...
IRJET - Automatic Lip Reading: Classification of Words and Phrases using Conv...IRJET - Automatic Lip Reading: Classification of Words and Phrases using Conv...
IRJET - Automatic Lip Reading: Classification of Words and Phrases using Conv...
 
Applications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and DesignApplications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and Design
 
GPT and other Text Transformers: Black Swans and Stochastic Parrots
GPT and other Text Transformers:  Black Swans and Stochastic ParrotsGPT and other Text Transformers:  Black Swans and Stochastic Parrots
GPT and other Text Transformers: Black Swans and Stochastic Parrots
 
XAI LANGUAGE TUTOR - A XAI-BASED LANGUAGE LEARNING CHATBOT USING ONTOLOGY AND...
XAI LANGUAGE TUTOR - A XAI-BASED LANGUAGE LEARNING CHATBOT USING ONTOLOGY AND...XAI LANGUAGE TUTOR - A XAI-BASED LANGUAGE LEARNING CHATBOT USING ONTOLOGY AND...
XAI LANGUAGE TUTOR - A XAI-BASED LANGUAGE LEARNING CHATBOT USING ONTOLOGY AND...
 
[DSC MENA 24] Nada_GabAllah_-_Advancement_in_NLP_and_Text_Analytics.pptx
[DSC MENA 24] Nada_GabAllah_-_Advancement_in_NLP_and_Text_Analytics.pptx[DSC MENA 24] Nada_GabAllah_-_Advancement_in_NLP_and_Text_Analytics.pptx
[DSC MENA 24] Nada_GabAllah_-_Advancement_in_NLP_and_Text_Analytics.pptx
 
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
APznzaalselifJKjGQdTCA51cF7bldYdFMvDcshM8opKFZ_ZaIV-dqkiLoIKIfhz2tS6Fw5UBk25u...
 
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
A NOVEL APPROACH FOR NAMED ENTITY RECOGNITION ON HINDI LANGUAGE USING RESIDUA...
 

Recently uploaded

Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
Joaquim Jorge
 

Recently uploaded (20)

Mastering MySQL Database Architecture: Deep Dive into MySQL Shell and MySQL R...
Mastering MySQL Database Architecture: Deep Dive into MySQL Shell and MySQL R...Mastering MySQL Database Architecture: Deep Dive into MySQL Shell and MySQL R...
Mastering MySQL Database Architecture: Deep Dive into MySQL Shell and MySQL R...
 
Tech Trends Report 2024 Future Today Institute.pdf
Tech Trends Report 2024 Future Today Institute.pdfTech Trends Report 2024 Future Today Institute.pdf
Tech Trends Report 2024 Future Today Institute.pdf
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
 
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
04-2024-HHUG-Sales-and-Marketing-Alignment.pptx
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
 
Data Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt RobisonData Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt Robison
 
Advantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your BusinessAdvantages of Hiring UIUX Design Service Providers for Your Business
Advantages of Hiring UIUX Design Service Providers for Your Business
 
Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024
 
Real Time Object Detection Using Open CV
Real Time Object Detection Using Open CVReal Time Object Detection Using Open CV
Real Time Object Detection Using Open CV
 
GenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day PresentationGenCyber Cyber Security Day Presentation
GenCyber Cyber Security Day Presentation
 
🐬 The future of MySQL is Postgres 🐘
🐬  The future of MySQL is Postgres   🐘🐬  The future of MySQL is Postgres   🐘
🐬 The future of MySQL is Postgres 🐘
 
HTML Injection Attacks: Impact and Mitigation Strategies
HTML Injection Attacks: Impact and Mitigation StrategiesHTML Injection Attacks: Impact and Mitigation Strategies
HTML Injection Attacks: Impact and Mitigation Strategies
 
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemkeProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
 
Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024Finology Group – Insurtech Innovation Award 2024
Finology Group – Insurtech Innovation Award 2024
 
Partners Life - Insurer Innovation Award 2024
Partners Life - Insurer Innovation Award 2024Partners Life - Insurer Innovation Award 2024
Partners Life - Insurer Innovation Award 2024
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
Developing An App To Navigate The Roads of Brazil
Developing An App To Navigate The Roads of BrazilDeveloping An App To Navigate The Roads of Brazil
Developing An App To Navigate The Roads of Brazil
 
Boost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityBoost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivity
 
Automating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps ScriptAutomating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps Script
 

Nlp 2020 global ai conf -jeff_shomaker_final

  • 1. Presentation at Global AI Conference Santa Clara Convention Center Santa Clara, CA January 23, 2020 Jeff Shomaker Founder, 21 SP, Inc. Natural Language Processing (NLP): Future Trends
  • 2. 21 SP, Inc. Proprietary and Confidential 2 Introduction • Natural Language Processing (NLP) is making rapid progress • Talk will cover: – A little history – What is NLP and what can you do with it? – Popular NLP Models – Major Model Advancements in 2019 – Top 10 AI Use Cases by Revenue – Commercial Products Announced in 2019 – Future Trends for NLP
  • 3. 21 SP, Inc. Proprietary and Confidential How New is NLP? – In 1637, Descartes wrote that language model-based machines would likely be possible. He said: For one may conceive that a machine would be so made that it could generate speech…but not so that it could adapt its output in a sufficiently versatile way so as to answer the meaning of everything that could be said in its presence, as even the stupidest of man can. 1) (1) Francois Chollet (2019 Nov 19). @fchollet, Tweet, See Bibliography for full citation. 3
  • 4. 21 SP, Inc. Proprietary and Confidential What Can NLP Do? – NLP can be used for the following: • Question answering • Speech recognition • Text-to-speech and speech-to-text • Topic modeling • Sentiment classification • Language modeling • Translation • Others. 2) (2) Rachael Thomas (2019 May-Jun). fast.ai Code-First Intro to Natural Language Processing, youtube, accessed 11-11-19. See bibliography. 4
  • 5. 21 SP, Inc. Proprietary and Confidential NLP Models – In the past, machine learning-based NLP processing was done with support vector machines (SVM) and logistic regression. – Recently, neural network (NN) models have been increasingly used and have provided better results on some NLP tasks. – Deep learning NNs use a “… multi-level automatic feature representation….” approach. – In the following, we will discuss four types of models used for NLP: • Convolutional neural networks (CNNs) • Recurrent neural networks (RNNs) • Recursive neural networks (Recur NNs) 3) • Transformer models. 17) (3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography. (17) Ashish Vaswani, et al (2017 Dec 6). Attention is all you need [On Transformers]. See bibliography. 5
  • 6. 21 SP, Inc. Proprietary and Confidential NLP: Convolutional Neural Networks (CNNs) – Collobert and Weston proposed CNNs for sentence modeling and other NLP tasks in 2008 and 2011 respectively. – Word embeddings - an approach that identifies which words are similar and the relationships between them. – After word embeddings came about, it was necessary to extract “… higher-level features from constituting words or n-grams.” 20) – NLP models use these for translation, question answering, summarization and sentiment analysis. – CNNs are good for capturing semantic information. – CNNs are not so good when data scarcity exists and they struggle when modeling some contextual information. 3) (3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See Bibliography. (20) N-grams are a sequence of words, eg, a 2-gram might be “please turn”. S. Kapadia (2019 Mar 26). Language Models: N-Gram, See bibliography. . 6
  • 7. 21 SP, Inc. Proprietary and Confidential NLP: Recurrent Neural Networks (RNN) – RNN design (ie, sequential information processing) makes them good for NLP since they do the same thing on each instance of the sequence. 3) – Since RNNs process in a looping fashion and they reuse weights, information is remembered during processing. 4) – The process works well for language modeling, machine translation, speech recognition and image capturing. – It is not clear whether RNNs or CNNs are best for sentiment classification, question answering or part-of-speech-tagging. – One drawback with simple RNNs is they struggle in learning and tuning an NN’s early layers. LSTMs (ie, RNN variant) overcome this issue. 3) (3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography. (4) Vincent Boucher (2019 Nov 25). AI Overview, academy.montreal.ai. See bibliography. 7
  • 8. 21 SP, Inc. Proprietary and Confidential NLP: Recursive Neural Networks (Recur NNs) – Recursive NNs use a tree-structure to account for the fact that in language “… words and sub-phrases combine into phrases in a hierarchical manner.” – Recur NNs work well for parsing* and there are two types: • Recursive autoencoders – can reconstruct context • Recursive neural tensor network – at each node in a tree, they compute a supervised objective. 3) – Recur NNs can be used for the following applications: • NLP • Audio-to-text transcription • Image scene decomposition. 5) * Parsing – “Analyze (a string or text) into logical syntactic components, typically in order to test conformability to a logical grammar.” LEXICO, lexico.com. (3) Tom Young, et al (2018 Nov 25). Deep learning trends for NLP. See bibliography. (5) J. Patterson & A. Gibson (2017). Deep Learning: A Practitioner’s Approach. See bibliography. 8
  • 9. 21 SP, Inc. Proprietary and Confidential NLP: Transformers – The Transformer model is a new approach to processing text that changed thinking and enabled Google’s BERT, which set the standard. 16), 18) – The Transformer uses only attention mechanisms, which allow them to avoid recurrence and convolutions. – This new architecture makes it possible to process in a parallel fashion, which can speed up training significantly over RNNs. – In this paper, which proposed this innovative architecture, Transformers were tested on translation tasks using BLEU: • For English-to-German, results showed improvement over all other models • For English-to-French, results bested all earlier single models. 17) (16) Prateek Joshi (2019 Jun 19). How do Transformers Work in NLP? See bibliography. (17) Ashish Vaswani, et al (2017 Dec 6). Attention is all you need [On Transformers]. See bibliography. (18) Jacob Devlin, et al (2019 May 24 – 1st version 2018 Oct 11). BERT: Pre-training of Deep Bidirectional Transformers. See bibliography. On 11 NLP tasks, BERT beat all earlier models. 9
  • 10. 21 SP, Inc. Proprietary and Confidential 2019 Advancement: XLNET – XLNET is a “... Generalized autoregressive pretraining method ….” that allows learning in a bidirectional manner and has improved performance compared to BERT. – XLNET has achieved excellent results on: • Language understanding • Reading comprehension • Test classification • Document ranking. – Using many benchmarks, XLNET is superior to BERT. 11) – Potential business applications include: • Automated customer support • Sentiment analysis (eg, brand awareness) • Information searching in document data bases. 12) (11) Zhilin Yang, et al (2019 Jun 19). XLNET: Generalized Autoregressive Pretraining. See bibliography. (12) Mariya Yao (2019 Nov 12). Major NLP Achievements in 2019. See bibliography. 10
  • 11. 21 SP, Inc. Proprietary and Confidential 2019 Advancement: ALBERT – A Lite BERT – ALBERT is an architecture that uses a transformer encoder.* – ALBERT scales better than BERT due to parameter reductions and an improved method for predicting sentences. – With fewer parameters, memory needs are reduced and training speed is increased. – ALBERT delivers state-of-the-art performances on GLUE, RACE, and SQuAD tests. 15) – Potential business applications include: • Chatbots • Sentiment analysis • Document mining • Text classification. 12) *Transformers are a major improvement over RNN-based sequence-to-sequence models used previously. (15) Zhenzhong Lan, et al (2019 Oct 30). ALBERT: A Lite BERT for Self-Supervised Learning. See bibliography. (12) Mariya Yao (2019 Nov 12). Major NLP Achievements in 2019. See bibliography. 11
  • 12. 21 SP, Inc. Proprietary and Confidential 2019 Advancement: DistilBERT – For the Edge – Transfer learning is used more and more in NLP; but, these big models are hard to train to do inference on-the-edge (eg, small devices). 21) – These large models can have parameters in the hundreds of millions. – Paper’s authors propose an approach for a “… general purpose language representation model .…” that can run on-the-edge. – DistilBERT use knowledge distillation which “… is a compression technique in which a compact model – the student – is trained to reproduce the behavior of a larger model – the teacher – or an ensemble of models.” – They discuss models that are: • 40% smaller than BERT • Keep 97% of BERT’s language understanding • Are 60% faster than BERT. 13) (13) Victor Sanh, et al (2019 Oct 16). DistilBERT, A distilled version of BERT. See bibliography. (21) Dipanjan (DJ) Sarkar (2018 Nov 14). A Comprehensive Hands-on Guide to Transfer Learning. See bibliography. Transfer learning allows information gained from an earlier model to be used for training a subsequent model (eg, reuse weights). 12
  • 13. 21 SP, Inc. Proprietary and Confidential AI Revenue -Top Ten Use Cases Forecast 6) 13 (6) Mike Quindazzi (2019 Dec 7). Cumulative AI Software Revenue, @tractica via @mikequindazzi, Tweet, See bibliography.
  • 14. 21 SP, Inc. Proprietary and Confidential 2019 New Product: Qualcomm’s Chipset – High end Android phones will soon have a new chipset called the Snapdragon 865. – The chip will make possible 5G, digital drivers licenses, a 200 megapixel camera and improved gaming on smartphones. – A very notable feature is “Instant live translation and transcription.” – Google Translate already has very good quality -- Android 10 provides live video transcriptions and Pixel 4 does translations with it’s voice recorder app. – It is reported that the demonstrations with the 865 are far superior. They show: • On-the-fly translation in English and Chinese • Words spoken in English can be translated and then spoken in Chinese to the listener. 9) (9) Jessica Dolcourt (2019 Dec 19). 6 Things Coming to Android Phones, CNET. See bibliography. 14
  • 15. 21 SP, Inc. Proprietary and Confidential 2019 New Product: Automatic Transcription – Amazon has brought out a product called Transcribe Medical • Uses a voice app and Amazon’s Web Services to automatically transcribe patient-doctor discussions • Once the discussion is complete, a transcription is placed in patient’s records. – Patient health records can be further analyzed by Amazon’s Comprehend Medical product • This analyzes text for patterns that can be used to highlight possible diagnoses and approaches to treatment. – Nuance and Microsoft collaborated earlier and deployed an improved Dragon Medical Virtual Assistant using Microsoft’s Azure platform. – Google is also developing a medical transcription product with Stanford University. 8) (8) Eric H. Schwartz (2019 Dec 6). Amazon Launches Medical Transcription, voicebot.ai. See bibliography. 15
  • 16. 21 SP, Inc. Proprietary and Confidential 2019 New Product: AI-enabled Robot w/ Emotion – CIMON (Crew Interactive Mobile Companion) was the first smart astronaut assistant used on the International Space Station (ISS). – CIMON is a collaboration between IBM (Watson AI), Airbus, and the German Aerospace Center. – The robotic assistant is designed to retrieve information, track tasks, and with the new version, CIMON-2, mitigate issues between crew members. – CIMON-2, which was sent aboard December 2019’s SpaceX ISS resupply mission, now uses IBM Watson’s Tone Analyzer. – The goal is to analyze astronaut’s conversations to allow for: • The mitigation of group-think • Provide an objective perspective or contrarian viewpoint when complicated decisions need to be made. 10) (10) D. Etherington (2019 Dec 5). AI-enabled Assistant Returns to Space, Techcrunch. See bibliography. 16
  • 17. 21 SP, Inc. Proprietary and Confidential A 2020 NLP Wish List – Sebastion Ruder’s list includes: • Learning from few samples rather than large datasets • Compact and efficient rather than huge models • Evaluate on at least another language (from a different language family) • New datasets contain at least one other language • NLP helps to unlock scientific knowledge (see Nature 19)). 7) (7) Sebastian Ruder (2019 Dec 23). 2020 Wish Lists, NLP News, Issue #48. See bibliography. (19) Vahe Tshitoyan, et al (2019 Jul 3). Unsupervised learning captures latent knowledge. See bibliography. 17
  • 18. 21 SP, Inc. Proprietary and Confidential 2020 AI and Machine Learning Trends – Natural Language Processing • Pre-training and fine tuning is the approach • Improvements in NLP are likely to be driven by the use of linguistics and the incorporation of more human knowledge • Neural machine translation (ie, simultaneous machine translation) is now practical and is expected to improve further – Conversational AI • Chatbots are widely used for customer service • To improve Chatbots, researchers are: – Leveraging conversational history and context – Working towards providing more varied responses – Attempting to incorporate emotion and empathy into conversations. 14) (14) Kate Koidan (2020 Jan 2). Top AI Trends for 2020, TopBots. See bibliography. 18
  • 19. 21 SP, Inc. Proprietary and Confidential End Notes • 1) Francois Chollet [Google Deep Learning] (2019 Nov 19), @fenollet, Tweet, accessed 12-18-19. • 2) Rachael Thomas (2019 May-Jun). fast.ai Code-First Intro to Natural Language Processing, What is NLP? (NLP Video 1), youtube, accessed 11-11-19. • 3) Tom Young, et al (2018 Nov 25). Recent Trends in Deep Learning Based Natural Language Processing, ar.Xiv:1708.02709v8 [cs.CL], accessed November 2019. • 4) Vincent Boucher (2019 Nov 25). Montreal.AI Academy: Artificial Intelligence 101 First World-Class Overview of AI for All, Montreal.AI, academy.montreal.ai, accessed 12-28-19. • 5) Josh Patterson and Adam Gibson, Deep Learning: A Practitioner’s Approach, O’Reilly 2017, accessed 12-22-19 at allite.books.com. • 6) Mike Quindazzi (2019 Dec 7). Cumulative Artificial Intelligence Revenue, Top Ten Use Cases World Markets:2017-2025, @tractica via @mikequindazzi, Tweet, accessed 12-29-19. • 7) Sebastian Ruder [Deep Mind AI] (2019 Dec 23). 2020 Wish Lists, Natural Language Processing News, Issue #48, newsletter.ruder.io, accessed 12-31-19. • 8) Eric Hal Schwartz (2019 Dec 6). Amazon Launches Medical Transcription Service in Direct Competition with Nuance [Nuance partner is Microsoft], voicebot.ai, accessed 12-30-19. • 9) Jessica Dolcourt (2019 Dec 12). 6 Things Coming to 2020 Android Phones: 8K Video, 5G Uploads, Two-finger Unlock, CNET, cnet.com, accessed 1-1-20. 19
  • 20. 21 SP, Inc. Proprietary and Confidential End Notes (cont.) • 10) Darrell Etherington (2019 Dec 5). AI-enabled Assistant Robot Returning to the Space Station with Improved Emotional Intelligence, Techcrunch, techcrunch.com, accessed 12-31-19. • 11) Zhilin Yang, et al (2019 Jun 19). XLNET: Generalized Autoregressive Pretraining for Language Understanding. arXiv:1906.08237v1 [cs.CL]. • 12) Mariya Yao (2019 Nov 12). What Are Major NLP Achievements & Papers from 2019, TopBots, topbots.com, accessed 12-31-19. • 13) Victor Sanh, et al (2019 Oct 16). DistilBERT, a Distilled Version of BERT: Smaller, Faster, Cheaper and Lighter, arXiv:1910.01108v2 [cs.CL], accessed 1-8-20. • 14) Kate Koidan (2020 Jan 2). Top AI & ML Research Trends for 2020, TopBots, topbots.com, accessed 1-5-20. • 15) Zhenzhong Lan, et al (2019 Oct 30). ALBERT: A Lite BERT for Self-Supervised Learning of Language Representations, arXiv:1909.11942v3 [cs.CL], accessed 1-2-20. • 16) Prateek Joshi (2019 Jun 19). How do Transformers Work in NLP: A Guide to the Latest State-of- the-Art Models, Analytics Vidhya, analyticsvidhya.com, accessed 1-10-20. • 17) Ashish Vaswani, et al (2017 Dec 6). Attention Is All You Need [Proposes The Transformer model], arXiv:1706.03762v5 [cs.CL]. • 18) Jacob Devlin, et al (2019 May 24 – 1st version 2018 Oct 11). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, arXiv:1810.04805v2 [cs.CL]. 20
  • 21. 21 SP, Inc. Proprietary and Confidential End Notes (cont.) • 19) Vahe Tshitoyan, et al (2019 Jul 3). Unsupervised Word Embeddings Capture Latent Knowledge from Materials Science Literature, Nature, nature.com, accessed 1-17-20. • 20) Shashank Kapadia (2019 Mar 26). Language Models: N-Gram, Towards Data Science, towardsdatascience.com, accessed 1-2-20. • 21) Dipanjan (DJ) Sarkar (2018 Nov 14). A Comprehensive Hands-on Guide to Transfer Learning with Real-World Applications in Deep Learning, Towards Data Science, towardsdatascience.com, accessed 1-19-20. • 21
  • 22. 21 SP, Inc. Proprietary and Confidential Contacts • Jeff Shomaker – Founder/President 21 SP, Inc. – jshomaker@21spinc.com – www.21spinc.com – 650-285-8122 • 21 SP, Inc. is a small privately held startup developing and marketing expert systems-based decision support software to use in genetic- based personalized medicine. The company's mission is to create tools that will reduce the use of traditional trial-and-error medicine by using pharmacogenetics and other evidence-based data, such as the results of high quality clinical trials, in the medical clinic. 22