Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
OpenTalks.AI - Сапунов Григорий, Ландшафт облачных когнитивных сервисов
Search
OpenTalks.AI
February 14, 2019
Science
0
540
OpenTalks.AI - Сапунов Григорий, Ландшафт облачных когнитивных сервисов
OpenTalks.AI
February 14, 2019
Tweet
Share
More Decks by OpenTalks.AI
See All by OpenTalks.AI
OpenTalks.AI - Виктор Лемпицкий, Моделирование 3Д сцен: новые подходы в 2020 году
opentalks
0
420
OpenTalks.AI - Алексей Чернявский, Нейросетевые алгоритмы для повышения качества медицинских изображений
opentalks
0
400
OpenTalks.AI - Александр Громов, Устойчивость нейросетевых моделей при анализе КТ/НДКТ-исследований
opentalks
0
350
OpenTalks.AI - Денис Тимонин, Megatron-LM: Обучение мультимиллиардных LMs при помощи техники Model Parallelism
opentalks
0
460
OpenTalks.AI - Егор Филимонов, Возможности платформы Huawei Atlas и эффективный гетерогенный инференс.
opentalks
0
110
OpenTalks.AI - Александр Прозоров, Референсная архитектура робота сервисного центра в отраслях с изменчивыми бизнес-процессами
opentalks
0
340
OpenTalks.AI - Наталья Лукашевич, Анализ тональности по отношению к компании — с чем не справился BERT
opentalks
0
320
OpenTalks.AI - Константин Воронцов, Фейковые новости и другие типы потенциально опасного дискурса: типология, подходы, датасеты, соревнования
opentalks
0
390
OpenTalks.AI - Дмитрий Ветров, Фрактальность функции потерь, эффект двойного спуска и степенные законы в глубинном обучении - фрагменты одной мозаики
opentalks
0
410
Other Decks in Science
See All in Science
DEIM2024 チュートリアル ~AWSで生成AIのRAGを使ったチャットボットを作ってみよう~
yamahiro
3
660
Machine Learning for Materials (Lecture 4)
aronwalsh
0
680
Onsager代数とその周辺 / Onsager algebra tsudoi
usamik26
0
390
データで課題を解決する -因果関係を調べる統計的因果推論-
sshimizu2006
4
1.3k
量子コンピュータとデータサイエンティスト
fuyu_quant0
0
130
Non-Gaussian methods for causal discovery
sshimizu2006
0
180
脳とAIは似ているか ― NeuroAI の挑戦
ykamit
9
6.8k
スポーツメトリクス設計に対比較法を使いまくる / Sports metrics design using pairwise comparison method (spoana#14)
konakalab
1
720
Direct Preference Optimization
zchenry
0
150
OptimizationNight~機械学習と数理最適化の融合~
hidenari
0
310
Mastering Feature Engineering: Mining the Hidden Salary Formula with CakeResume
tlyu0419
0
130
The Universe, How it works
arafkarsh
0
110
Featured
See All Featured
GraphQLとの向き合い方2022年版
quramy
33
12k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
6
3.4k
Rebuilding a faster, lazier Slack
samanthasiow
74
8.2k
The Brand Is Dead. Long Live the Brand.
mthomps
49
29k
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
245
20k
Imperfection Machines: The Place of Print at Facebook
scottboms
261
12k
Become a Pro
speakerdeck
PRO
12
4.6k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
358
22k
Design by the Numbers
sachag
274
18k
The MySQL Ecosystem @ GitHub 2015
samlambert
244
12k
Art, The Web, and Tiny UX
lynnandtonic
290
19k
Keith and Marios Guide to Fast Websites
keithpitt
408
22k
Transcript
Сloud cognitive services landscape Grigory Sapunov / CTO, Intento ©
Intento, Inc. February 2019
INTENTO 2 © Intento, Inc. / February 2019 A simple
API to third-party AI models from many vendors https://inten.to
AI SUPPLY SIDE STACK 3 Hardware and infrastructure Deep Learning
Frameworks Neural Network Architectures Training Inference Data Processing Data Data Collection API Management Domain Adaptation © Intento, Inc. / February 2019
4 Hardware and infrastructure Deep Learning Frameworks Neural Network Architectures
Training Inference Data Processing Data Data Collection API Management Domain Adaptation ON THE SHOULDERS OF GIANTS http://www.shivonzilis.com/machineintelligence © Intento, Inc. / February 2019
A PLENTY OF SUPPLY 5 APIs 200+ intents 100+ AI
models 20,000+ © Intento, Inc. / February 2019
6 23 intents 60+ APIs © Intento, Inc. / Confidential
/ June 2018 EXAMPLE: NLP APIS © Intento, Inc. / February 2019 6
7 © Intento, Inc. / Confidential / June 2018 EXAMPLE:
MT APIS 7 Based on Jan 2019 report on MT: https://bit.ly/stock_mt_jan2019
DOES IT WORK? SOMETIMES 8 of companies using AI get
ROI >30% 83% Deloitte State of Cognitive Survey 2017
DOES IT WORK? SOMETIMES 9 of them use third-party AI
80% of companies using AI get ROI >30% 83% Deloitte State of Cognitive Survey 2017
DOES IT WORK? NOT QUITE 10 of them use third-party
AI 80% of companies using AI get ROI >30% 83% Deloitte State of Cognitive Survey 2017 of all companies ever tried to use AI 20%
LEGACY (SAAS) BUYING PROCESS 11 Good brand — Set of
features — Price — Pilot Integration “AI IS NOT THERE YET!” Six months later: © Intento, Inc. / February 2019
WHAT MAKES THE AI DIFFERENT? 12 trained on vast amount
of unknown data — performance varies up to 4x case by case — constantly updated (not always improving) — has its own quirks — decisions often made by non engineers © Intento, Inc. / February 2019
FEATURES DIFFER SIGNIFICANTLY (SENTIMENT ANALYSIS EXAMPLE) 13 © Intento, Inc.
/ February 2019 13
QUALITY DIFFERS SIGNIFICANTLY (MACHINE TRANSLATION EXAMPLE) 14 Based on July
2018 report on MT: https://bit.ly/mt_jul2018
QUALITY CONSTANTLY CHANGES (MACHINE TRANSLATION EXAMPLE) 15 Based on July
2018 report on MT: https://bit.ly/mt_jul2018
ANOTHER RED FLAG: PRICE 16 up to 300x difference for
similar services — Machine Translation: $3 to $960 per 1M symbols — Sentiment Analysis: $1 to $99 per 10K API calls https://bit.ly/stock_mt_jan2019 https://bit.ly/sa_overview © Intento, Inc. / February 2019
EXAMPLE: SENTIMENT ANALYSIS PRICES 17 © Intento, Inc. / February
2019
NO SINGLE BEST/OPTIMAL VENDOR 18 Based on Jan 2019 report
on MT: https://bit.ly/stock_mt_jan2019
YOU CAN COMBINE SEVERAL VENDORS 19 Based on custom NMT
report: https://bit.ly/custom_nmt_nov2018
DEVELOPER EXPERIENCE (DX) 20 More info at https://bit.ly/intento_dx 20
IT’S MORE LIKE HIRING HUMANS 21 evaluate source onboard ©
Intento, Inc. / February 2019
WITH ITS OWN PROBLEMS 22 different APIs and data formats
— hard to evaluate — hard to use multiple vendors concurrently — hard to source © Intento, Inc. / February 2019
Programmable Web AI MODELS: SOURCING 23 Major AI Vendors AI
API Marketplaces Indie AI Vendors Deprecated AI APIs © Intento, Inc. / February 2019
AI MODELS: EVALUATION 24 lots of vendor reports and marketing
— web demos — … <the tools chasm> … — API integrations NOTHING! © Intento, Inc. / February 2019
AI MODELS: ONBOARDING 25 different APIs, SDKs, CLIs (changed and
deprecated from time to time) — connectors and plugins to client systems — domain adaptation © Intento, Inc. / February 2019
HOW INTENTO CHANGES THAT? 1. CORE TECHNOLOGY 26 we build
and maintain many AI API integrations — intent-based vendor-agnostic API — high-performance AI middleware solution — the same convenience tools for every API: SDK, CLI, developer console evaluate source onboard © Intento, Inc. / February 2019
HOW INTENTO CHANGES THAT? 2. SOURCING 27 Intento Service Catalog
https://inten.to/api-platform — monitoring changes in API specs, documentation and behavior — regular public evaluation reports (performance, price and feature comparison) evaluate source onboard © Intento, Inc. / February 2019
HOW INTENTO CHANGES THAT? 3. EVALUATION 28 Single API https://github.com/intento/intento-api
— human-friendly evaluation tools (command-line interface) and evaluation API — API request smart-routing, optimizing performance, price or latency evaluate source onboard © Intento, Inc. / February 2019
HOW INTENTO CHANGES THAT? 4. ONBOARDING 29 Single API, Vendor-agnostic
SDK and developer tools https://console.inten.to — Re-usable plugins and connectors to internal systems — No vendor-lock, can use multiple vendors simultaneously evaluate source onboard © Intento, Inc. / February 2019
ACCESS TO MANY MODELS EASY TO PICK THE RIGHT ONE
30 API MARKETPLACES INDIE VENDORS MAJOR VENDORS EASY TO CHOOSE HARD TO CHOOSE FEW MODELS MANY MODELS © Intento, Inc. / February 2019
CONCLUSIONS 31 lots of niche AI models with APIs —
complex performance / price landscape — AI users: choose AI portfolio, not a single vendor — AI vendors: - avoid inventing new APIs - make tools for humans as well © Intento, Inc. / February 2019
THANK YOU! 32 Grigory Sapunov
[email protected]
2150 Shattuck Ave Berkeley
CA 94705 https://inten.to
[email protected]