Professional Documents
Culture Documents
Cognitive Systems (Unit 5)
Cognitive Systems (Unit 5)
Cognitive Systems (Unit 5)
—-------------
UNIT 5
Introduction
Healthcare Ecosystem: The healthcare industry is vast and comprises various organizations
working together to support patient wellness and care.
Roles in the Ecosystem: There are several well-defined roles within the healthcare
ecosystem, including healthcare providers, payers, medical device manufacturers,
pharmaceutical firms, research labs, health information providers, and regulatory agencies.
Technological Advances: While there have been significant technological advancements in
healthcare, there's a pressing need for continued innovation to improve health outcomes for
patients.
Data Management Challenges: Healthcare data is often managed in silos, making it difficult
to share patient and medical research data across different stakeholders. The growing
volume and variety of healthcare data exacerbate this challenge.
Issues with Data Consistency and Connectivity: Even when organizations are willing to
share information, the data often lacks consistency and connectivity, hindering progress in
medical research and leading to clinical errors.
Impact of Medical Errors: Medical errors, whether preventable or not, contribute significantly
to mortality rates. Depending on the measurement methodology, preventable harm from
medical errors ranks as one of the leading causes of death.
—--------------------------------------------------
Data Variety in Healthcare: The healthcare industry generates and manages a vast amount
of data, including digital images, medical reports, patient records, clinical trial results, and
billing records. This data exists in various formats and systems, ranging from manual paper
records to structured and unstructured digital data.
Challenges in Data Management: The lack of integration among different systems poses
significant challenges for healthcare organizations in managing and analyzing the vast
amount of data generated. However, addressing these challenges presents opportunities for
improving health outcomes.
Role of Electronic Medical Record (EMR) Systems: Healthcare providers have implemented
EMR systems to maintain integrated, consistent, and accurate patient records. Despite being
a work in progress for many organizations, EMRs offer benefits such as facilitating confident
and speedy treatment decisions with access to complete and up-to-date patient information.
Importance of Finding Patterns and Outliers: Healthcare organizations face the persistent
challenge of identifying patterns and outliers in both structured and unstructured data to
improve patient care.
The shift is towards integrated knowledge bases that include structured and unstructured
data, moving away from document-centric silos.
Move Towards Standards-Based Approach: Healthcare data management is transitioning
towards a standards-based approach to facilitate data sharing. Medical devices and sensors
can generate valuable data about a patient's condition, offering opportunities for improving
patient screening and anticipating medical condition changes using predictive analytical
models.
Role of Cognitive Systems: Cognitive systems can capture and integrate sensor-based data
with the entire history of medical research and clinical outcomes to enable significant
improvements in outcomes. For example, analytical models developed by doctors in a
neonatal department provide advance warning on infants at risk of developing
life-threatening infections, leveraging real-time monitoring and data analysis.
—------------
Complex Data Sharing Dynamics: Each constituent in the healthcare ecosystem has access
to different sources of relevant healthcare data. While some data is shared, much of it is
controlled by regulations and security requirements. The relationships between the
constituents in terms of data sharing are complex and constantly evolving.
Providers: Providers manage a wide range of structured and unstructured data sources,
including patient medical records (EMR, doctors’ notes, lab data), sensor data, intake
records, medical textbooks, journal articles, clinical research studies, regulatory reports,
billing data, and operational expense data.
Data Service Providers: Data service providers offer prescription drug usage and
effectiveness data, healthcare terminology taxonomies, and software solutions for analyzing
healthcare data.
—----------------------------------------------------
Learning from patterns in Healthcare data
Benefit of Cognitive Computing: Healthcare professionals often deal with massive amounts
of data, ranging from patient records to medical research. Cognitive computing helps them
sift through this data efficiently, extracting valuable insights that aid in making informed
decisions. For example, it can analyze patient symptoms, medical history, and treatment
options to recommend the most effective course of action.
Importance of Accurate Data: Accurate, trusted, and consistent data forms the foundation of
effective healthcare decision-making. For example, when diagnosing a patient, physicians
rely on accurate medical histories, test results, and imaging scans to make informed
decisions about treatment plans. Cognitive computing ensures data accuracy by applying
rigorous analysis techniques and validation processes.
Challenges in Data Interpretation: Interpreting healthcare data can be challenging due to its
complexity and variability. For instance, patient symptoms may manifest differently across
individuals, making diagnosis and treatment decisions more nuanced. Cognitive computing
addresses these challenges by synthesizing diverse data sources, identifying patterns, and
generating actionable insights to support healthcare professionals in their decision-making
process.
Learning from Data Patterns: By analyzing patterns in healthcare data, cognitive systems
can identify trends and correlations that may not be immediately apparent. For instance, they
can detect early warning signs of potential health complications or predict patient outcomes
based on historical data. This predictive capability enables proactive interventions, ultimately
improving patient outcomes and reducing healthcare costs.
Predictive Models for Readmission Rates: Predictive models leverage historical patient data
to forecast the likelihood of hospital readmissions. By analyzing factors such as patient
demographics, medical history, and treatment outcomes, these models can identify patients
at higher risk of readmission. Healthcare providers can then implement targeted
interventions to mitigate these risks and improve patient care quality.
Early Stage Implementation: While there are exciting examples of cognitive systems in
healthcare, it's important to note that these implementations are still in their infancy. This
means that although there's promise, widespread adoption and maturity are still to come.
Privacy and Security: While leveraging data for analytics, healthcare organizations must
prioritize privacy and security. This involves ensuring that personal identifying information is
safeguarded and removed from datasets to comply with regulations and protect patient
confidentiality.
Genomic Data Analysis: The increasing volume of genomic data necessitates advanced
computational approaches for storage, processing, and analysis. Sophisticated algorithms
and tools play a crucial role in understanding this data, empowering scientists to tackle
complex biological questions with efficiency and accuracy.
—----------------------
Two Paths of Implementation: Cognitive healthcare applications are progressing along two
main paths: customer or user engagement applications and discovery applications.
Discovery Applications: These applications are utilised in scenarios like drug discovery or
determining the optimal treatment for patients. They leverage cognitive systems to
understand relationships and discover patterns in data to enhance healthcare outcomes.
Understanding User Types: It's crucial to understand the types of users accessing the
cognitive healthcare application and their medical background and expertise. This includes
determining if users are medical students, experienced clinicians, or health and wellness
consumers.
Incorporating Changes: User requirements and expectations evolve over time, necessitating
ongoing adjustments to the cognitive system's development. Continuous learning ensures
the system becomes more intelligent and valuable to end users with increased usage.
—----------------
Purpose of Ontologies: Ontologies map relationships between terms with similar meanings,
aiding in the organization and understanding of medical terminology.
Examples of Ontologies: Various ontologies are widely used in healthcare, covering medical
conditions, treatments, diagnostic tests, drug information, and complications. One example
is the International Classification of Diseases (ICD), with ICD-10 being a commonly used
version.
—--------------------
Starting with User Questions: Before building the knowledge base, gather the types of
questions users will ask.
Application Strategy: Define the overall application strategy before reviewing data sources.
Risks of Starting with Corpus: Beginning with the corpus may lead to targeting questions to
the assembled sources, potentially missing user needs.
Representation of User Questions: Initial questions should reflect the various types of
queries users will make.
Importance of Question Quality: Getting the questions right is crucial for the future
performance of the application.
Seeding the Cognitive System: Begin with a sufficient number of question-answer pairs to
initiate the machine learning process.
Quantity of Question-Answer Pairs: Typically, 1000-2000 pairs are needed to kickstart the
learning process.
Voice of End User: Questions should be in the voice of the end user, while answers should
be provided by subject matter experts.
Corpus Definition: The corpus is like the brain of the cognitive application—it holds all the
knowledge needed for the system to function. Just like how your brain stores information to
help you think and make decisions, the corpus stores all the documents and data that the
cognitive system uses to answer questions and provide responses.
Inclusion of Documents: Imagine the corpus as a library where all the books contain
information that the cognitive system can access. Every document that the system needs to
read and understand is included in this library. This ensures that the system has access to
all the information it requires to provide accurate answers and responses.
Content Collection Process: Think of the content collection process as gathering all the
necessary books for the library. The question-answer pairs act as a guide, helping to decide
which books (or documents) are needed to provide the right information. This process
ensures that the library (corpus) has all the books (content) it needs to function effectively.
Identifying Required Content: Before stocking the library, it's essential to know what books
are needed. Similarly, in building the corpus, it's crucial to identify the types of information
required to answer questions accurately. This includes resources like medical texts, clinical
studies, and patient records, which provide the necessary knowledge base for the cognitive
system.
Examples of Content: The content included in the corpus spans a wide range of sources,
from medical textbooks and research papers to patient records and ontologies. These
diverse sources ensure that the cognitive system has access to comprehensive information
relevant to healthcare and medical inquiries.
Validation of Content: Just like how you check that books in a library are readable and
understandable before adding them to the shelves, content for the corpus undergoes
validation. This ensures that the information is clear and comprehensible, meeting the
standards required for inclusion in the cognitive system's knowledge base.
Meta Tags and Associations: Meta tags act as labels on books, helping to categorize and
organize them in the library. Similarly, adding meta tags to content assists in creating
associations between documents, facilitating easier navigation and retrieval of information by
the cognitive system.
Content Formatting: Proper formatting of content ensures that it's presented in a way that the
cognitive system can easily understand. This includes organizing content into sections and
headings and optimizing formats for clarity and comprehension.
Source Transformation: Sometimes, the information needed for the corpus might be in a
complex format, like a table. To make it easier for the cognitive system to process, these
complex formats may need to be transformed into simpler, unstructured text.
Scheduled Updates: Just like how libraries periodically add new books and replace old ones
to keep their collection up to date, the corpus needs regular updates to ensure that it
remains current and reflects the latest information in the healthcare domain.
Continuous Update Process: Establishing a process for continuous updates ensures that the
corpus evolves alongside new developments in healthcare. This ongoing update process is
vital to maintaining the effectiveness and relevance of the cognitive system over time.
Importance of the Patient: The patient, or healthcare consumer, is at the core of the
healthcare ecosystem. Their health and well-being generate extensive data crucial for
healthcare management.
Benefits of Health Improvement: Improving health outcomes, such as weight loss, increased
exercise, balanced diet, and quitting smoking, leads to significant benefits for healthcare
payers, governments, and organizations.
Difficulty in Making Positive Changes: Despite knowing the risks, many people find it
challenging to make positive lifestyle changes to improve their health.
—------
Welltook
CaféWell Concierge: CaféWell Concierge serves as a central hub for users to access
personalized health resources and programs. It organizes various tools and programs, such
as tracking devices and apps, to create tailored plans based on each user's unique needs
and preferences.
Partnership with IBM Watson: By partnering with IBM Watson, Welltok incorporates
advanced cognitive technologies into CaféWell Concierge. This enables the app to
understand user queries, analyze data, and deliver personalized guidance and
recommendations using natural language processing, machine learning, and analytics.
Incentives and Rewards: To encourage user engagement, health payers provide incentives,
such as gift cards or premium reductions, for using CaféWell Concierge. Advanced analytics
algorithms ensure that these incentives are aligned with user actions, motivating individuals
to adopt and maintain healthy behaviors.
Training Process: Welltok gathers user input to generate question/answer pairs, which are
reviewed by experts to ensure accuracy and relevance. These pairs form the basis for
training the application, which continuously enhances its response quality by leveraging
Watson's cognitive capabilities and refining its knowledge base.
—------
GenieMD
Mobile Application Accessibility: GenieMD's services are available to users via a mobile
application, enabling easy access to health-related information and support on-the-go. This
mobile platform enhances user convenience and facilitates continuous engagement with
health resources and recommendations.
Improved Health Outcomes: GenieMD strives to enhance health outcomes for patients by
empowering them to actively manage their health and make informed decisions. By
promoting patient engagement and providing personalized guidance, the company aims to
achieve positive health outcomes while potentially lowering healthcare expenses.
—-----
—-------
—---------------------
—--------------------------
TRAVEL –
Example: WayBlazer:
WayBlazer, founded by Terry Jones, aims to revolutionize the travel industry by harnessing
the power of cognitive computing.
The company partners with IBM Watson to utilize its cognitive computing services, including
natural language processing and hypothesis generation/evaluation capabilities.
WayBlazer's goal is to provide evidence-based advice and personalized recommendations
to travelers by analyzing vast amounts of data from destination and transportation vendors.
By collaborating with organizations like the Austin Convention and Visitor’s Bureau,
WayBlazer seeks to create customized travel applications that cater to individual preferences
and needs.
Furthermore, WayBlazer plans to expand its services to offer concierge services to hotels
and airlines, thereby enhancing the overall user experience and generating additional
revenue opportunities.
Infrastructure Safety:
Ensuring the safety of transportation infrastructure is a critical and ongoing concern for
companies in the industry.
Investments in security measures and technologies are essential to protect against potential
threats and mitigate risks to operations and personnel.
—-------
Hypothesis Formation: The process begins with creating a hypothesis to address a specific
problem or question.
Data Gathering: All relevant data pertaining to the problem area is collected and aggregated.
Data Vetting and Cleansing: The gathered data is vetted to ensure reliability and accuracy.
It's then cleaned and verified to remove inconsistencies or errors.
Training the Data: The data is utilized to train the system, enhancing its ability to understand
and process information effectively.
Application of NLP and Visualization: Techniques like natural language processing (NLP)
and visualization are applied to interpret and present the data in a comprehensible format.
Refinement of the Corpus: The dataset undergoes refinement to improve its quality and
relevance to the problem being addressed.
Continuous Analysis with Predictive Analytics: Once operational, the system continuously
analyzes the data using predictive analytics algorithms to identify trends and patterns.
Iterative Process: The entire cycle is iterative, with insights gained from analysis informing
the refinement of hypotheses and approaches, initiating the process anew.
Creation of a Dynamic Learning Environment: This life cycle fosters a sophisticated and
dynamic learning environment, enabling ongoing improvement and adaptation.
—------------------------------
Additional Interfaces:
Depending on the task, additional interfaces beyond NLP may be necessary.
Visualization interfaces can help researchers identify patterns visually, aiding tasks like drug
development.
Voice recognition technology improvements allow systems to detect emotions like fear or
confusion, enhancing user guidance.
BabyX Project:
BabyX is an experimental virtual infant prototype developed at the University of Auckland's
Laboratory for Animate Technologies.
It combines bioengineering, computational neuroscience, artificial intelligence, and
interactive computer graphics research.
BabyX is designed to simulate basic neural systems involved in interactive behavior and
learning, analyzing real-time video and audio inputs to react to caregivers' or peers'
behavior.
The project aims to advance understanding and development of computational models for
interactive systems, incorporating advanced 3D computer graphics and behavioral modeling.
Continuous Development:
BabyX is continuously evolving, with ongoing development in neural models, sensing
systems, and real-time computer graphics realism.
It represents a cutting-edge exploration of the potential of visual interfaces and
computational models for interactive systems.
—-------------
Level of Transparency:
Users can modify and adapt subsets of the package to suit their unique needs.
Some packages may become industry standards, serving as ubiquitous best practices.
—----------------
Well-Defined Services:
There will be a wide range of well-defined services available for tasks such as data
ingestion, real-time analysis, and data visualization.
Natural language interfaces will empower users to interact with these services in a more
intuitive manner.
Rather than presenting raw data, there will be a shift towards delivering narratives that
provide meaningful insights and explanations.
This shift towards narrative-based interfaces will enhance understanding and
decision-making for users across various industries.
—--------------
Permission-Based Interactions:
Permission-based interactions will remain essential, but more automated techniques will
emerge.
Systems will analyze patterns across millions of interactions to determine appropriate
actions and security levels.
The optimal system will operate in the background, suggesting actions when necessary and
respecting user preferences.
—-----------------------
EMERGING INNOVATIONS
Deep QA Overview: Deep QA, which stands for Deep Question Answering, is a method
where a system generates probing questions for humans to answer. However, it's not
commonly practiced outside certain contexts. One notable example is IBM's Watson, which
uses Deep QA interactively with experts, particularly in complex fields like healthcare. Here's
how it works: Imagine a scenario where a doctor describes symptoms of a patient. Watson
then steps in, asking specific questions to refine the potential diagnoses. By doing this,
Watson helps narrow down the possible answers and increases the confidence in
diagnosing the patient's condition.
Tracking Information: In the realm of Deep QA, it's crucial for systems to keep track of the
information provided during a session. This means they need to remember what has been
discussed previously. The system should only ask further questions when the human's
answers can actually improve its own performance. So, it evaluates the potential answers
and assigns confidence levels to them. Then, it considers whether there's enough evidence
to make a decision or if more information is necessary. This approach ensures that the
system maintains continuity and makes informed decisions based on the available data.
Knowledge Sharing: One of the exciting aspects of Deep QA is the potential for shared
learning experiences among different systems that tackle related questions. This
collaboration can lead to the development of reusable patterns across various domains.
Take healthcare, for instance. By aggregating Deep QA analysis, researchers and
practitioners can discover optimal treatments for specific conditions, such as skin cancer.
This collaborative effort allows systems to accumulate knowledge over time and streamline
the problem-solving process as both data and analysis mature.
Future Trends: Looking ahead, it's likely that problem-solving methods, especially in complex
domains, will shift towards cognitive computing. Similar to how the scientific method guides
discoveries in natural sciences, Deep QA and hypothesis generation will become standard
approaches in various professional fields. This means that the future of problem-solving will
involve asking probing questions, generating hypotheses, and testing them to find optimal
solutions. As this approach becomes more prevalent, it will shape the way problems are
tackled across different disciplines.
2. NLP
Advances in NLP: NLP has made significant strides recently, exemplified by IBM's Watson's
ability to understand complex text even under challenging conditions, like the Jeopardy!
format, where answers may be ambiguous. Watson's success in this format highlights its
capability to grasp meaning effectively, even when faced with tricky language nuances or
uncertainties.
Key NLP Innovation: A crucial innovation in NLP would involve identifying common
underlying structures among languages and emulating the manual processes used by expert
translators to discover rules or heuristics unconsciously applied. By analyzing well-respected
translations of texts, NLP systems can gain insights into commonalities and differences,
enabling them to develop more robust translation algorithms.
Hardware innovations play a crucial role in shaping the evolution of cognitive computing both
in the short and long term. Currently, traditional hardware systems, primarily von Neumann
architecture computers, are used to build cognitive systems. These systems rely on parallel
structures, but the actual processing occurs within central processing units (CPUs) or
adjunct processors like graphical processing units (GPUs). However, major breakthroughs
are expected in chip architectures and programming models in the near future.
—------------------------------
Use of Grammatical Rules: NLP techniques rely on the grammatical rules of a language to
understand the meaning of words, phrases, sentences, or entire documents. These rules
help in recognizing predictable patterns within the language.
Inference Making: NLP applies known rules and patterns to infer meaning from text
documents. By analyzing the structure and content of the text, it can make educated
guesses about the intended meaning.
Identification and Extraction of Elements: NLP techniques can identify and extract various
elements of meaning from text, such as proper names, locations, actions, or events. This
helps in understanding relationships among different elements, even across multiple
documents.
Application in Database Analysis: NLP techniques are not limited to analyzing standalone
text documents but can also be applied to text within databases. For example, they can be
used to find duplicate names and addresses or analyze comments or reason fields in large
customer databases.
—-----------------------------------
Task of NLP: NLP's primary goal is to translate unstructured text into a meaningful
knowledge base. It helps make sense of messy text by organizing it into a format that users
can understand and interact with.
Linguistic Analysis: NLP breaks down text to extract meaning and enable users to ask
questions and receive relevant answers. It dissects language to understand its components
and structure, making it easier for users to interact with the text and obtain useful
information.
Importance of Context: Understanding context is crucial for NLP, as it helps in assessing the
true meaning of text-based data by identifying patterns and relationships between words and
phrases. NLP relies on context to decipher the intended meaning behind words and phrases,
ensuring accurate interpretation and analysis of text.
Applications: NLP is utilized in various scenarios, such as trip planning for a truck driver or
medical data review for a lung cancer specialist, to assist users in making informed
decisions based on textual information. It is employed in real-world situations to support
decision-making processes by extracting valuable insights from text, catering to specific user
needs and requirements.
Complexities of Language: Language poses challenges due to its ambiguity and multiple
meanings, which NLP tools address by interpreting language and extracting essential
elements. NLP tools navigate the complexities of language by deciphering ambiguous words
and phrases, ensuring accurate comprehension and analysis of text content.
Understanding Context: NLP starts with basic elements like words and builds up context by
identifying parts of speech, references, and relationships between entities. It establishes
context by analyzing the structure and content of text, enabling it to identify key elements
and their relationships within the context of the document or conversation.
—-------------------
Subjectivity and Interpretation: The same words or sentences can carry different meanings
depending on the context, speaker, audience, and cultural factors. Humans often use
language subjectively, infusing words with personal experiences, emotions, and intentions.
This subjectivity leads to varied interpretations of language.
Truth Stretching and Manipulation: In communication, individuals may stretch the truth or
manipulate words to convey specific meanings or achieve desired outcomes. This
manipulation of language adds layers of complexity to understanding communication and
requires careful interpretation of context.
—---------------------
Understanding Linguistics
Modeling Natural Languages: NLP focuses on modeling the structure and patterns of natural
languages to interpret their meaning automatically. This involves analyzing both the
grammatical and semantic aspects of language to capture its underlying rules and
conventions.
Grammatical and Semantic Patterns: NLP seeks to uncover the grammatical and semantic
patterns that occur within languages or specific sublanguages, such as those used in
specialized fields like medicine or law. These patterns help in understanding how words and
phrases are structured and interpreted within different contexts.
Levels of Meaning: NLP examines various levels of meaning to enhance understanding. This
includes considering not only the surface meaning of words and sentences but also the
deeper semantic nuances and connotations that contribute to our comprehension.
—--------------
In any analysis of incoming text, the fi rst process is to identify which language the text is
written in and then to separate the string of characters into words (tokenization ). Many
languages do not separate words with spaces, so this initial step is necessary.
—---------------------
Phonology
Definition of Phonology: Phonology is the study of the physical sounds of a language and
how those sounds are produced or pronounced within that language. It focuses on
understanding the patterns and rules governing the sounds used in speech.
Importance for Speech Recognition and Synthesis: Phonology plays a crucial role in speech
recognition and synthesis systems. By studying the phonetic characteristics of a language,
these systems can accurately interpret and generate spoken language.
Limited Relevance to Written Text: Phonology is not as important for interpreting written text
since written language does not directly involve the production or perception of sounds.
Instead, written text relies on visual symbols (e.g., letters, characters) to convey meaning.
Intonation and Emotion: Intonation patterns, including variations in pitch, stress, and rhythm,
convey important information about emotions and attitudes. For example, differences in
intonation can indicate whether a speaker is angry, confused, excited, or sad, even if they
use the same words.
—------------
Morphology
Definition of Morphology: Morphology refers to the structure of words, including their stems
and additional elements that convey meaning. It involves analyzing whether a word is
singular or plural, its tense, and other grammatical features.
Partitioning into Morphemes: Words are broken down into smaller units called morphemes,
which help determine their meaning. These morphemes include prefixes, suffixes, infixes,
and circumfixes, each adding a specific element of meaning to the word.
Role of Lexicon and Grammar Rules: A lexicon or repository of words and grammar rules
assists in interpreting meaning. Techniques like parts-of-speech tagging and tokenization
help identify words with definitive meanings, especially in specialized industries like
medicine.
Contextual Significance: Certain terms have specific meanings within particular industries or
disciplines. For example, "blood pressure" has a distinct meaning in medicine, highlighting
the importance of context in language interpretation.
—------------------
Syntax Analysis
Syntax refers to the rules that govern how sentences are structured in languages.
Understanding both the syntax and semantics of natural language is crucial for cognitive
systems to deduce meaning based on the context in which language is used. Words can
have different meanings depending on the specific industry or context in which they are
used, leading to ambiguity. Syntactical analysis, or parsing, involves analyzing the
arrangement of words in a sentence according to grammar rules. This process helps
systems understand the meaning of language in context and is vital for tasks like
question-answering, where accurate parsing ensures accurate responses. For instance,
parsing correctly in a question like "Which books were written by British women authors
before 1800?" ensures that the system focuses on identifying books rather than authors.
—----------------
—--------