We assume that we are perfectly healthy, most of times we don’t know what’s going on inside our bodies, until the day we travel to the hospital in pain or agony, only to discover a serious or life-threatening condition. The chances are that the ailment didn’t start that morning; rather, it’s been growing or developing for some time. The patient was simply never aware of it. At that point, once doctor diagnosed as “sick,” our medical system engages to take care of you. What if, instead of this retrospective and reactive approach, we were constantly monitored, so that we could know the moment anything was out of whack? Better yet, what if we more closely monitored those aspects of our body that our gene sequence predicted might cause us difficulty? Think: our heart, our kidney, our breasts. Such a system becomes personalized, predictive and possibly preventative. The same is possible to build Applications enabled by AI, ML, big data and in future quantum computers and other powerful technologies like 3d printing. Machine learning and drug design Artificial Intelligence and Big Data in medicine Healthcare and AI
2. One of most exciting time in history
Zero
•Initial
condition
i.e.13.7 billion
years back
•Big bang
started with
pure energy
•Elementary
particles
1st Novelty
•Between 380K
year after big
bang
•Formation of first
atom i.e. H2
followed by
other stable
elements of
periodic table
formed either in
core of star or
during star-burst
by nuclear fusion
2nd Novelty
•10 billion years after
big bang
•Life from organic
molecules started
i.e. single cell life
followed by multi-
cellular life
3rd Novelty
•200-300K before
present time
•Life became
intelligent and
achieved higher
consciousness
primarily by us i.e.
homo sapiens
4th Novelty
•Started i.e.
current time
•When natural
intelligence is
extending its
power by artificial
intelligence
•Likely to lead to
technology
singularity in few
decades
•May achieve
higher
consciousness
with augmented
intelligence
4. Training
• Used to Create Model
• Leverages Training
Data
• Throughput-intensive
Two Main Phases in ML
Inference
• Feed Input to Model
• Used for prediction
• Latency-sensitive
• Can be done at Edge
Inference
Big Data / Analytics
internetofmedicalthings/Edge
Machine
Learning
Deep
Learning
Training
Training
Training
Machine Learning
5. 5
Machine Learning Pipelines are Complex
MLConfiguration
Data Collection
Feature Extraction
Data
Verification
Machine
Resource
Management
Analysis Tools
Process
Management Tools
Serving
Infrastructure Monitoring
Source: Hidden Technical Debt in Machine Learning Systems, Scully, D., et al (Google)
END-TO-END ML:
CORE ML ALGORITHMS AND SUBSTANTIAL ”SURROUNDING” INFRASTRUCTURE
IT domain
Data Scientists/ Engineer
Domain
Users / Healthcare
Specialists
Input/Requirements
8. AI Transforming healthcare
AI, ML
& Big
Data
Keeping
healthy
Early
detection
Enhanced
diagnosis
Augmented
decision
making
Proactive
treatment
Elderly Care
Drug
research &
discovery
CRISPR
Microbiome
Neural
Science
9. How can I collect
my data, and
maintain privacy
What health
measurements I
can do at home ?
What
smart
wearables
should
I
Use?
What
supplements
should I take ?
Is intermittent
fasting good for
me?
Shall I do my
Genome
profiling ?
Sickcare vs Healthcare