ithaka the next wave 2016: john rinderle - acrobatiq
TRANSCRIPT
John RinderleCTO, acrobatiq
LINEAGE
“Can we apply what we know about how people
learn to measurably improve
learning outcomes online?”
LINEAGE
< 50 hours~18% learning
gain
Adaptive, Data-Driven OLI
Course
Traditional College Course
> 100 hours~3% learning
gain
Replicated 3x at CMU
External report by ITHAKA
Result 1: Accelerated Learning
Result 2: Increased Retention
Adaptive/Accelerated group scored higher than Traditional Control, p < .05.
Result 3: Improved Transfer
LINEAGENumbers
12 year project | 104 faculty & 55 institutions
$20,000,000 from NSF, Gates, Hewlett…
1,250,000 independent learners
150,000+ academic students
Results 20+ published papers
learning, time
3 segments: Private; 4 Yr. Public; Community Colleges
LINEAGENumbers
12 year project | 104 faculty & 55 institutions
$20,000,000 from NSF, Gates, Hewlett…
1,250,000 independent learners
150,000+ academic students
Results 20+ published papers
learning, time
3 segments: Private; 4 Yr. Public; Community Colleges
Acrobatiq created to sustainably scale the effective practices of OLI as a platform.
Acrobatiq Platform
“Smart” AuthoringApplication
Adaptive Learning Environment
Learning Analytics
Smart Author
Organize work into projects. Invite teams. Customize roles and workflow.
Smart Author
Intuitive editor with built in instructional design guidance.
Smart Author
Out of the box templates for a variety of content, interaction and question types.
Smart Author
Skill graph or knowledge map that defines and measures your goals for students.
Adaptive Learning PlatformClear Goals Learn by Doing Practice w/ Targeted Feedback Synthesize & Apply
Learning Science-Based Design
Predictive ModelingCognitive Factors
++
Content Knowledge Model Learning Estimates
Student Data
From Data Points to InsightsMillions of data points are modeled against course outcomes or goals to predict level of learning for each.
13
Ripples of Data• Adapt or inform the student
• Fewest data points, most local intervention
• Inform the instructor• More data, still fast results
• Administrator• Shape program design
• Content creator• Target course improvement cost effectively
• Learning scientist• Improve our understand of human learning• Most data, slowest impact
Student A has high learning estimates in Module 20, and is directed to the key question.
Student B has low learning estimates, and is scaffolded to the same key questionvia the presentation of questions and hints tied to skills they are deficient on.
The Learning Dashboard organizes participation and learning data to quickly give insight into student performance. Instructors customize
views to their particular needs…
… the most powerful views combine learning and participation to provide better insight and guide instructors to the most helpful interventions.
Data feeds back course designer to guide refinement and improvement.
Big Data and Machine Learning
Three Opportunities
1. Focus Support Services Where Needed• Better insights helps scale
student support services:• Identify at-risk students• Intervene earlier• Efficient coaching sessions• Recommend actions to
learners and instructors• Just-in-time remediation
and extra practice• Match learners for help• Direct help to point of
greatest need• Target course and concept
“hotspots”
Attain Bachelor’s Degree by Age 24
1970 2013
Top Income Quartile
40% 77%
Bottom Income Quartile
6% 9%
University of Pennsylvania’s Alliance for Higher Education and Democracy and the Pell Institute for the Study of Opportunity in Higher Education.
Unblocking “Clogged” ArteriesSmall targeted improvements can make big differences in student outcomes.
Autopilot?
Can machine learning enable a course to “drive” itself?
Autopilot – Wrong Goal?The stakes for learners are high!• At-risk students require support services that are difficult to automate.• Predictive models must be accurate:
Underestimating erodes confidence. Overestimating sets up failure.
Virtual Assistant
Ask questions to better understand studentprogress and performance. Drill in to find answers.
Lane Departure Warning
Alert course designers, instructors, students if they are in danger of going off path.
Lane Departure Warning
Alert course designers, instructors, students if they are in danger of going off path.
Blind Spot Detection
Detect trends and patterns and surface previously unknown insights.
Blind Spot Detection
Detect trends and patterns and surface previously unknown insights.
Blind Spot Detection
Detect trends and patterns and surface previously unknown insights.
Emergency “Braking”Connect students to the help they need, before they ”crash”. Equip advisors withinformation to make interventions more timely and effective.
2. Why and what to do about it?New techniques combine machine learning AND learning science to enable inferences which explain problems and suggest actionable solutions.
2. Why and what to do about it?New techniques combine machine learning AND learning science to enable inferences which explain problems and suggest actionable solutions.
3. Better Utilize Existing Digital Resources
• University libraries hold vast collections of licensed, public domain, and open content
• NLP to discover and recommend applicable content
• Share faculty authored content
• Efficacy data about what content helps most
When additional help or practice is needed, where does the content come from?