Hacc 2023 Google Ai - ML Workshop

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 59

HACC 2023

Google AI/ML
Workshop

Daniel Liu
[email protected]
Google AI Mission Statement

Google AI is focused on bringing the


benefits of AI to everyone.
We do this through research that advances the
state-of-the-art in the field, efforts to apply AI
to Google products and to new domains, and
by developing tools to ensure that everyone
can access AI.
“Machine learning is a core,
transformative way by which we’re
rethinking how we’re doing everything.”

- Sundar Pichai, Google, 2016


What is
AI?
AI is a bigger concept to create intelligent
machines that can simulate human
thinking capability and behavior
Machine Learning is a specific field of
AI where a system learns to find
patterns in examples in order to make
predictions.
Computers learning how to do a task
without being explicitly programmed
to do so.
Explicitly Programmed
(Flowchart)
Machine Learning Allows You to Solve a Problem
Without Codifying the Solution

✓ Recognizes patterns in data

✓ Predictive analytics at scale

✓ Builds ML models seamlessly

✓ Fully managed service


Google Cloud AI
✓ Deep Learning capabilities

Proprietary + Confidential
Google Cloud End-to End AI Platform
Accelerate Business Outcomes with Enterprise-Ready Machine Learning Pipeline

Fraud Targeted Recommendation Predictive


Detection marketing Engine Analytics
Industry Use-cases
In-loop inferencing for trained models Predictive
Risk Customer
Demand Forecast Inventory
Analysis Segmentation
Management

Cloud AI products
Pre-trained ML APIs to
Building custom ML models

ML Framework
Industry-standard & widely adopted

Infrastructure
Best-in class processors for ML/DL
CPU GPU TPU

Proprietary + Confidential
Confidential + Proprietary
“If everyone spoke to their
phone for three minutes,
we’d exhaust all available
computing resources”

Jeff Dean
Google Senior Fellow
2014

Confidential & Proprietary


Global
Datasphere
Survey by IDC
● IDC defines the "global datasphere" as "the quantification of the amount of
data created, captured, and replicated across the world."

Google Cloud TPU

Proprietary + Confidential
TPU vs
Conventional
processors
15 - 30x faster
30 - 80x operations per watt

Like fast-forwarding 7 years


into the future
AI can be complex and time
intensive

Data ML model Tune ML Evaluate Deploy Update


preprocessing design model
parameters

Large computational Machine learning Manual data


resource expertise labeling
End to End: Google Cloud AI Spectrum

Cloud ML Perception services


MLE

Use/extend OSS SDK Build custom models Use pre-built models

ML researcher Data Scientist App Developer

Proprietary + Confidential
Cloud AI products & solutions
Services & Solutions Collaboration Services
CxOs
Solutions
Ease of
Implementation
Talent Contact Document Profession Cloud AI
Center AI Understanding AI AI Hub ASL al Services Partners
Solution
Sight Language Conversation Structured Data
APIs
Video Natural Speech- Text-to- Dialogflow Recommendations AI
Vision Language Translation
Intelligence to-Text Speech Enterprise
Building
Blocks

Sight Language Structured Data


AutoML
Builders

Natural
Vision Video Translation Tables
Language

Built-in Tools On-prem Integrated with


AI Platform
New New New New
Platform

Data Pre-built Notebook VM Images Data


Development Datasets Training Predictions Kubeflow
Dataflow Dataproc
BigQuery Dataprep Studio
Environment Labeling Algorithms

Accelerators Frameworks
Infrastructure
AI Foundation
TPU GPU CPU
Cloud AI products & solutions
Services Solutions Collaboration Services
CxOs
&
Solutions
Contact Document Profession Cloud AI
Ease of Talent
Center AI Understanding AI AI Hub ASL
Solution al Services Partners
Implementatio
n
Sight Language Conversation Structured Data
APIs
Video Natural Speech- Text-to- Dialogflow Recommendations AI
Vision Language Translation
Intelligence to-Text Speech Enterprise
Building
Blocks

Sight Language Structured Data


AutoML
Builders

Natural
Vision Video Translation Tables
Language

Built-in Tools On-prem Integrated with


AI Platform
Development New New New New
Platform

Environment Data
Datasets Data Kubeflow
Notebook VM Images Training Predictions Studio
Dataflow Dataproc BigQuery Dataprep

Accelerators
Pre-built Labeling Frameworks
Algorithms
Infrastructure
AI Foundation
TPU GPU CPU
Google is the pioneer in Bard
AI 202
3
A conversational
202 AI Service
2
AlphaFold predicts
powered by
2020 structures of all LaMDA.

Google LaMDA known proteins


201 Model Trained to
9
Text-to-Text converse
201 Transfer Transformer
8
Google’s LLM 10B P Model
201 groundbreaking Open Sourced
7
Google invents large language
201 Transformer model, BERT 3,000 Researchers
6 7,000 Publications
2015 Google’s DeepMind kickstarting LLM
revolution
helps detect eye
disease Responsible AI
Google DeepMind
AlphaGo defeats Go
champion Upholds high scientific
Built & Tested for Safety Privacy in design
standards

Avoid creating unfair


Accountable to People Socially Beneficial
bias
Expanding our portfolio
To support the needs of Generative AI centric enterprise
development

Powered by Foundation Models Powered by Foundation Models Powered by Foundation Models Powered by Foundation Models

CCAI Document AI Discovery AI Healthcare AI


Business users
Conversation AI

Generative AI
App Builder Foundation Vertex AI Search
Models (Enterprise Search)

Developers

End-to-End ML Platform

Generative AI Studio | Generative AI APIs | Model


Vertex Garden
AI
AI Practitioners
Google Cloud Infrastructure
Confiden tial + Proprietary
Highlight A few Google AI
Solutions
1. Enterprise Translation Hub (ETH)

2. Cloud Video Intelligence and


Vision

3. Document AI

4. Contact Center AI (CCAI)

5. Generative AI and LLM


Google Cloud
Enterprise Translation
Hub
Google Translation Hub
Enterprise-Ready Self-Serve Personal Translator

ETH brings automated doc translation directly to users

● Self-serve ease-of-use, convenience & velocity


● Instant translation to >100 languages
● Document format preservation

Strong enterprise administration & control


Upload your doc, select target languages…

● Simple, transparent, per-page pricing


● Strong data security & access controls
● Access & deploy advanced features org-wide -
custom models, glossaries, human review

… and get a great translation in a few seconds


Google Translation Hub
Enterprise-Ready Self-Serve Personal
Translator

Get started today with translation tools known and loved by billions of users, thoughtfully connected
together in a Self-Serve platform built for the enterprise

Customized Translations Rich Layout Easy


Retention adoption
Demo
Google Cloud
Enterprise Translation
Hub Demo
Cloud Video
Intelligence and Vision
Using Google Cloud, Enables Memphis to detect potholes
the City of Memphis and vacant properties with over
90% accuracy
applies AI & ML to its
toughest public Projected 75% increase in number of
potholes identified, saving the city up to
works and urban $20,000 a year
planning problems
Improves residents' lives and
visitors' experiences with safer
streets and communities
Google
DocAI
Most business transactions
begin, involve or end with a
document
Document AI enables you to unlock insights
from your documents with machine learning
Document AI extracts & classifies information
from unstructured documents

0 02 03
1
Read Understand it Make it
it useful
Which unlocks significant
value

Operational efficiency Customer experience Insights


Rea
d
● OCR (Optical Character Recognition)
Document AI
approaches documents
like people do
Understand
● Natural language
● AutoML Natural Language
Natural Language
API
Classify, analyse & extract information about
people, places, events, and more

● Multilingual support Classify content Detect


● Extract key document entities sentiment
● Analyze sentiment

Extract entities Analyze


syntax
AutoML Natural Language: Sentiment analysis
Understand the overall attitudes expressed based on domain-specific sentiment
scores

prediction
results

-0.7 0.9
Life of a
document CATEGORIZE CONTENT
Categorize patent’s content
DETECT DIAGRAMS
INGEST & FILTER Identify diagram and
Patent is read in from Cloud using NLP model. corresponding x & y
Storage. Non-patents are coordinates.
Meet Patent:
filtered out.
Unstructured
document,
multiple formats
& languages

OCR
Extract out raw text into STORAGE
json format for Write out and store results
downstream NLP process. EXTRACT ENTITIES from the pipeline into
Identify named entities in BigQuery
the raw text.
READ : OPTICAL CHARACTER
RECOGNITION US010114351B2
(12) United States Patent
Fadell et al.
(10) Patent No.: US
10,114,351 B2
(45) Date of Patent: Oct. 30,
2018
(54)
(56)
References Cited
SMART-HOME AUTOMATION SYSTEM
THAT SUGGESTS OR AUTMATICALLY
IMPLEMENTS SELECTED
HOUSEHOLD POLICIES BASED ON
SENSED OBSERVATIONS
U.S. PATENT DOCUMENTS
4,475,685 A *
10/1984 Grimado ..............
F23N 5/203
236/46 R
(71)
Applicant:
GOOGLE
INC.,
Mountain
View, CA
(US)
7,689,920 B2
9,330,274 B2
9,450,962 B2
*
3/2010 Robbin et al.
5/2016 Schepis et al.
9/2016 Longhorn ......
(Continued)
H04L 43/50
(72)
FOREIGN PATENT
DOCUMENTS
Inventors: Anthony M. Fadell, San Francisco, CA
(US); Yoky Matsuoka, Palo Alto, CA
CLASSIFY CONTENT
US010114351B2
(12) United States Patent
Fadell et al.
(10) Patent No.: US
10,114,351 B2
(45) Date of Patent: Oct. 30,
2018
Categorization Confidence
(54)
(56)
References Cited
SMART-HOME AUTOMATION SYSTEM
Computer Vision 0.961
THAT SUGGESTS OR AUTMATICALLY
IMPLEMENTS SELECTED
Med Tech 0.030
HOUSEHOLD POLICIES BASED ON
SENSED OBSERVATIONS
Cryptocurrencies 0.009
U.S. PATENT DOCUMENTS
4,475,685 A *
10/1984 Grimado ..............
F23N 5/203
236/46 R
(71)
Applicant:
GOOGLE
INC.,
Mountain
View, CA
(US)
7,689,920 B2
9,330,274 B2
9,450,962 B2
*
3/2010 Robbin et al.
5/2016 Schepis et al.
9/2016 Longhorn ......
(Continued)
H04L 43/50
(72)
FOREIGN PATENT
DOCUMENTS
Inventors: Anthony M. Fadell, San Francisco, CA
(US); Yoky Matsuoka, Palo Alto, CA
(US); David Sloo, Menlo Park, CA
(US); Maxime Veron, Los Altos, CA
EXTRACT ENTITIES
US010114351B2 Publication date: Oct. 30, 2018
(12) United States Patent
Fadell et al. Classification_1: G05B 15/02
(10) Patent No.: US
10,114,351 B2 Classification_2: G05B 15/02
(45) Date of Patent: Oct. 30,
2018 Application Number: 114,351
(54)
(56)
SMART-HOME AUTOMATION Filing Date: MAR. 5, 2015
References
SYSTEM Cited
Applicant: GOOGLE INC.
THAT SUGGESTS OR AUTMATICALLY
IMPLEMENTS SELECTED Inventor: Returns inventor name
HOUSEHOLD POLICIES BASED ON
SENSED OBSERVATIONS First Line of Patent Title: SMART-HOME AUTOMATION
U.S. PATENT DOCUMENTS
4,475,685 A * SYSTEM
10/1984 Grimado ..............
F23N 5/203
236/46 R
(71)
Applicant:
GOOGLE
INC.,
Mountain
View, CA
(US)
7,689,920 B2
9,330,274 B2
9,450,962 B2
*
3/2010 Robbin et al.
5/2016 Schepis et al.
9/2016 Longhorn ......
(Continued)
H04L 43/50
(72)
FOREIGN PATENT
DOCUMENTS
Inventors: Anthony M. Fadell, San Francisco, CA
(US); Yoky Matsuoka, Palo Alto, CA
(US); David Sloo, Menlo Park, CA
(US); Maxime Veron, Los Altos, CA
DETECT DIAGRAMS

Diagram: Returns x, y coordinates of bounding boxes


STORE DATA
Document
AI
● Data
Document Identification
Storage
?
Case ● Logs
Management Information Parsing
System ● Business
Knowledge
eFile Upload system
Validation of Rules
● APIs
Hawaii Safe
Travels
Application
Application Overview
Google
DocAI Demo
https://
cloud.google.com/
document-ai/docs/drag-
and-drop
Contact Center
AI CCAI
CCAI automates simple interactions and enables
agents to solve issues quickly, using industry-leading
AI
1 Virtual Agent
Gives patients 24/7 access to immediate conversational self-service, with
seamless handoffs to live agents for more complex issues.

2 Agent Assist
Empowers agents with continuous support during their calls by
identifying intent and providing real-time, step-by-step assistance.

3 Insights
Insights Uses natural language processing to identify call drivers, popular questions,
and other information that helps contact center managers learn about patient
interactions to improve call outcomes.
CCAI Demo
Department of
Motor
Vehicles
(DMV)
Proprietary + Confidential

Use Cases
● Vehicle
Registration
Renewal
● Identity
Verification
● Credit Card
Payment
● Drive Test
Scheduling
● Agent
Assist

Demo URL
https://youtu.be/j8Y4q
PgR-C0

^Contains state
specific information
Generative AI and
LLM
Google is the pioneer in Bard
AI 202
3
A conversational
202 AI Service
2
AlphaFold predicts
powered by
2020 structures of all LaMDA.

Google LaMDA known proteins


201 Model Trained to
9
Text-to-Text converse
201 Transfer Transformer
8
Google’s LLM 10B P Model
201 groundbreaking Open Sourced
7
Google invents large language
201 Transformer model, BERT 3,000 Researchers
6 7,000 Publications
2015 Google’s DeepMind kickstarting LLM
revolution
helps detect eye
disease Responsible AI
Google DeepMind
AlphaGo defeats Go
champion Upholds high scientific
Built & Tested for Safety Privacy in design
standards

Avoid creating unfair


Accountable to People Socially Beneficial
bias
Consumers & enterprises have different needs….

Consumers Enterprises

Create a How do we
How do we deal with
Plan a 3 day valentine control our fraud &
trip to poem. data
Patagonia
security
How to make GF
pancakes?
We need to be How will we
accurate & control
A picture of a
panda playing explainable costs?
I want to
yahtzee write a novel.
Create a jazz
How do I get
song for a bday
started? How do we integrate our
card
existing data &
applications

Bard + Vertex
MakerSuite AI
Your Data, Your Terms
Your OnPrem or
Alternative Cloud

Your Google Cloud CMEK DRZ AxT VPC-SC

Perimeter
Generative AI App
Builder

Your Content
Conversation Vertex
AI AI Google Cloud Central
Chatbot, API, etc. Search
Your Data Hosting
Internet/Intranet

Your Inference Vertex AI

Large Base Model


Adapter
(Frozen)
Your
Layers
Your Users Security
Expanding our portfolio
To support the needs of Generative AI centric enterprise
development

Powered by Foundation Models Powered by Foundation Models Powered by Foundation Models Powered by Foundation Models

CCAI Document AI Discovery AI Healthcare AI


Business users
Conversation AI

Generative AI
App Builder Foundation Vertex AI Search
Models (Enterprise Search)

Developers

End-to-End ML Platform

Generative AI Studio | Generative AI APIs | Model


Vertex Garden
AI
AI Practitioners
Google Cloud Infrastructure
Confiden tial + Proprietary
Google Cloud Foundation
Models
Across a variety of model sizes to address use cases

PaLM for Text PaLM for Chat Imagen for Text to


Custom language tasks Multi-turn conversations with Image
session context Create and edit images from
simple prompts

Embeddings API Chirp for Codey for


for Text and Image Speech to Code Generation
Extract semantic information Text Improve coding and
from unstructured data Build voice enabled debugging
applications
The future of
customer
experience
Enhanced conversational capabilities and easier chatbot development
Section 04
Search

LLMs vs.
virtual agents
Conversation
LLMs are characterized by emergent abilities, or
the ability to perform tasks that were not included
in their training examples.

LLMs contextual understanding of human


language changes how we interact with data
and intelligent systems.
Content generation

LLMs can find patterns and connections in


massive, disparate data corpora.
Thank
You
Thank
you

You might also like