Emergence of Mathematical Abilities from Experience in Distributed Neural Networks Jay McClelland and the PDP lab at Stanford.

Slides:



Advertisements
Similar presentations
Summer 2011 Tuesday, 8/ No supposition seems to me more natural than that there is no process in the brain correlated with associating or with.
Advertisements

Experiments and Variables
Transformations We want to be able to make changes to the image larger/smaller rotate move This can be efficiently achieved through mathematical operations.
Mathematics Algebra 1 Trigonometry Geometry The Meaning of Numbers Choosing the Correct Equation(s) Trig and Geometry Mathematics is a language. It is.
Key Stone Problems… Key Stone Problems… next Set 11 © 2007 Herbert I. Gross.
Key Stone Problem… Key Stone Problem… next Set 21 © 2007 Herbert I. Gross.
Mathematics as a Second Language Mathematics as a Second Language Mathematics as a Second Language Developed by Herb I. Gross and Richard A. Medeiros ©
PD1: Getting started.
EXPERIMENTAL ERRORS AND DATA ANALYSIS
TIER ONE INSTRUCTION Comparing Fractions. Tier I Instruction Tier I is the highly effective, culturally responsive, evidence-based core or universal instruction,
Status of Middle School Mathematics Teaching 2000 National Survey of Science and Mathematics Education Dawayne Whittington Horizon Research, Inc.
Meaningful Learning in an Information Age
THE TRANSITION FROM ARITHMETIC TO ALGEBRA: WHAT WE KNOW AND WHAT WE DO NOT KNOW (Some ways of asking questions about this transition)‏
The Game of Algebra or The Other Side of Arithmetic The Game of Algebra or The Other Side of Arithmetic © 2007 Herbert I. Gross by Herbert I. Gross & Richard.
Dyscalculia Dyslexia Teaching Assistant Course December 2010.
Algebra Problems… Solutions
M ATH C OMMITTEE Mathematical Shifts Mathematical Practices.
Created by The School District of Lee County, CSDC in conjunction with Cindy Harrison, Adams 12 Five Star Schools SETTING GOALS (OBJECTIVES) & PROVIDING.
Key Stone Problem… Key Stone Problem… next Set 23 © 2007 Herbert I. Gross.
College Algebra Prerequisite Topics Review
Chapter 3: Equations and Inequations This chapter begins on page 126.
Arithmetic of Positive Integer Exponents © Math As A Second Language All Rights Reserved next #10 Taking the Fear out of Math 2 8 × 2 4.
Jay McClelland Stanford University
Copyright © Cengage Learning. All rights reserved.
Instructional Shifts for Mathematics. achievethecore.org 2 Instructional Shifts in Mathematics 1.Focus: Focus strongly where the Standards focus. 2.Coherence:
Middle School Liaison Meeting
Unit 9 Project Preview and Algebra By Jessica Rodriguez.
Incorporating the process standards into the daily rigor Incorporating the process standards into the daily rigor.
Three Shifts of the Alaska Mathematics Standards.
Common Core State Standards for Mathematics: Review Focus and Coherence A Closer look at Rigor.
K-12 Mathematics Common Core State Standards. Take 5 minutes to read the Introduction. Popcorn out one thing that is confirmed for you.
© Witzel, 2008 A Few Math Ideas Brad Witzel, PhD Winthrop University.
Dynamics of learning: A case study Jay McClelland Stanford University.
Development of Mathematical and Physical Reasoning Abilities Jay McClelland.
Transitioning to the Common Core: MDTP Written Response Items Bruce Arnold, MDTP Director California Mathematics Council – South Conference November 2,
Classroom Assessments Checklists, Rating Scales, and Rubrics
SOL Changes and Preparation A parent presentation.
Unultiplying Whole Numbers © Math As A Second Language All Rights Reserved next #5 Taking the Fear out of Math 81 ÷ 9 Division.
 The most intelligent device - “Human Brain”.  The machine that revolutionized the whole world – “computer”.  Inefficiencies of the computer has lead.
MM212 Unit 2 Seminar Agenda Simplifying Algebraic Expressions Solving Linear Equations and Formulas.
Representation of Symbolic Expressions in Mathematics Jay McClelland Kevin Mickey Stanford University.
Learning about Learning.... A Talk prepared for G-TEAMS and HEATWAVES GK-12 Project Fellows Debra Tomanek, Ph.D. Associate Vice Provost, Instruction &
Dynamic Presentation of Key Concepts Module 5 – Part 1 Fundamentals of Operational Amplifiers Filename: DPKC_Mod05_Part01.ppt.
Multiplying Whole Numbers © Math As A Second Language All Rights Reserved next #5 Taking the Fear out of Math 9 × 9 81 Single Digit Multiplication.
Algebra Form and Function by McCallum Connally Hughes-Hallett et al. Copyright 2010 by John Wiley & Sons. All rights reserved. 3.1 Solving Equations Section.
Emergence of Semantic Knowledge from Experience Jay McClelland Stanford University.
Transfer and Problems Solving Denise Nichols and Brant Kenny.
Solving Equations with Fractions. 2 Example: Solve for a. The LCD is 4. Simplify. Add 2a to both sides. Divide both sides by 3. Check your answer in the.
National Math Panel Final report 2008 presented by Stanislaus County Office of Education November 2008.
Algebra Problems… Solutions Algebra Problems… Solutions © 2007 Herbert I. Gross Set 10 By Herbert I. Gross and Richard A. Medeiros next.
Similarity and Attribution Contrasting Approaches To Semantic Knowledge Representation and Inference Jay McClelland Stanford University.
Teaching to the Big Ideas K - 3. Getting to 20 You are on a number line. You can jump however you want as long as you always take the same size jump.
Create a 5 Whys. Think about the purpose of maths and what type of mathematical learners you wish to create in the classroom.
Development of Expertise. Expertise We are very good (perhaps even expert) at many things: - driving - reading - writing - talking What are some other.
The Emergent Structure of Semantic Knowledge
Inductive and Deductive Reasoning  The pre-requisites for this chapter have not been seen since grade 7 (factoring, line constructions,..);
Carol Dweck (Stanford University) Adapted from How do people’s beliefs influence their motivation and subsequent achievement in academic.
Welcome to Math 6 Our subject for today is… Divisibility.
Emergent Semantics: Meaning and Metaphor Jay McClelland Department of Psychology and Center for Mind, Brain, and Computation Stanford University.
Chemistry Math Crunch Do you have what it takes?.
Balancing on Three Legs: The Tension Between Aligning to Standards, Predicting High-Stakes Outcomes, and Being Sensitive to Growth Julie Alonzo, Joe Nese,
Sum and Difference Formulas. WARM-UP The expressions sin (A + B) and cos (A + B) occur frequently enough in math that it is necessary to find expressions.
Week 1 Real Numbers and Their Properties (Section 1.6, 1.7, 1.8)
MATH BY MEAGHAN, ROWEN, ELSIE. CONTENT LIST ▪ INTRODUCTION : Past vs Present ▪ SELECTING APPROPRIATE MATH : Math Standards ▪ RESEARCH ON MATH INSTRUCTION.
Parent Introduction to Eureka Math
Presented by: Angela J. Williams
Software Quality Engineering
Emergence of Semantics from Experience
CHAPTER I. of EVOLUTIONARY ROBOTICS Stefano Nolfi and Dario Floreano
Toward a Great Class Project: Discussion of Stoianov & Zorzi’s Numerosity Model Psych 209 – 2019 Feb 14, 2019.
Presentation transcript:

Emergence of Mathematical Abilities from Experience in Distributed Neural Networks Jay McClelland and the PDP lab at Stanford

Why is Math so Hard to Learn? Late grade-school-aged kids misunderstand equations – What goes in the blank: = __ + 4 Many middle-school-aged kids misunderstand fractions – Is 19/20 closer to 1 or 21? Most Stanford undergraduates don’t understand the rudiments of trigonometry – Which expression below has the same value as cos(-30°)? sin(30°) -sin(30°) cos(30°) -cos(30°)

Failure to attach the appropriate meaning to mathematical expressions A fraction N/D represents a certain number N of pieces of a unit whole divided into D equal parts An equation represents an equivalence relation between two quantities, one to the left and one to the right of the equals sign The sine / cosine of an angle θ in degrees represents – the projection of a point on the unit circle specified by θ onto the vertical / horizontal axis through the center of the circle, – or equivalently, the coordinates of the point on the circle XXX 47 5?

cos(70)

cos(–70+0)

sin(-θ) cos(-θ) Reported Circle Use: “A Lot” “A Little” or “Not at all”

Who is to blame for these failures? The teacher / the textbook: – Too much emphasis on abstract concepts, rote procedures, and algebraic manipulation – Not enough emphasis on maintaining contact with the meaning of the concepts in question The students / their parents / our implicit theories about our abilities Yes all this is true… but still – the concepts seem very simple once you understand them – and they are being presented. So, Again, Why are they so hard to learn??

Habits of Mind 1 Learning to encode expressions automatically so that their meaning is readily apparent in the mind depends on a gradual strengthening process that occurs incrementally over repeated opportunities to learn – This is no different in principle from learning to read words aloud, or many other things we learn We quickly loose awareness that we are engaging in these processes – once they have been well practiced, the meaning of an expression comes to mind without explicit thought and appears to be intuitive and obvious. Margolis, H. (1987). Patterns, thinking and Cognition. U. of Chicago Press.

Can studies of learning in neural networks help dig more deeply into these issues? Example 1: – Learning to read Example 2: – Learning to represent numerosity Example 3: – Learning to solve equation problems Discussion and future directions

Neural Network Models of Representation and Learning Connections are real-valued, so representation and learning are real-valued also Connection-based knowledge can approximate discrete rule- like behavior, and can capture influence of continuous variables too Connection adjustment occurs via small increments, making change occur gradually Performance generally changes gradually, but can exhibit accelerations and decelerations. H I N T /h/ /i/ /n/ /t/

Warning: Simulation vs Theory The models I will describe deliberately simplifies a complex system by considering only some of its parts and by trying to extract key properties of learning systems in the brain rather than mimicking all of their details

FIND OWN FIVE TAKE RIND SOWN HIVE, HINT HAKE HIGH LOW FREQUENCY NETWORK ERROR REACTION TIME

2 3 4 HS GRADE MEAN ERRORS (out of 20)

Memorization, Rules or ?? Networks like this can generalize – they are not strictly memorizing their inputs Some earlier versions did not generalize as well as human subjects do, but other versions generalize quite well. For example, in Plaut et al 1996, the reading model read nonwords as well as human subjects do, and made a similar pattern of responses. – GAKE almost always pronunced to rhyme with TAKE – MAVE sometimes rhymes the SAVE, sometime with HAVE

Model’s Improvement With Experience RIND HAKE HAVE TAKE

Summary Connections strengthen gradually with experience; speed and accuracy of processing gradually increases The knowledge acquired generalizes: The network can read pronounceable nonwords as human subjects do Frequent and typical items are learned most quickly Less frequent items and less typical items are harder to learn, but are eventually mastered by the network The knowledge is implicit and becomes more and more robust and sensitive to complexities with experience

The Approximate Number System (ANS) Piazza et al. 2004

Progressive Improvement in Judging Numerosity and Area (Odic et al, 2013)

Stoianov & Zorzi (2013)

Progressive development of a representation that supports numerosity judgments At several points in training, the network is tested for it’s ability to use the representation At the top layer to judge whether the number of items in the input is greater or less than a standard

Results at Different Time Points

Children vs. Network Scaled Network ‘Age’

Summary Learning to do a non-numeric task can create a representation sensitive to numerosity in a very generic neural network Characteristics of biological numerosity can arise without the task of representing number per se The structure of the training set may matter for this – What factors are characteristic of natural experience? – What factors affect the network’s numerosity representations? Take-home point is that human-like sensistivity to number can arise and can be progressively refined from a very general architecture and learning mechanism

A neural network model that learns “the concept of equivalence” Or at least, it learns to pass behavioral tests whose success has led others to attribute implicit knowledge of the concept of equivalence A project by one of my PhD students, Kevin Mickey

Phenomena to be addressed Children answer incorrectly in problems of the form: a = b + __ They tend to put the sum of a and b in the blank, rather than the correct answer, which is b – a. When given such equations in a brief presentation, and asked to reproduce them, they tend to reproduce them as a + b = __ While the expressions used in studies are often more complex, these simple examples capture the essence of the phenomenon.

Analysis of Input Researchers have studied textbooks used in different school systems, and they find: – Operands are predominantly on the left of the equal sign in early-grade texts and examples ~90% of cases have operands only on the left – When a blank occurs it is by itself about 60% of the time – Thus, there are cases like __ + b = c or a + __ = c – But few very few cases like a = __ + c or a = b + __ Our training set mirrored these statistics

Important Point The statistics are stationary throughout the simulation – So the changing pattern in the network is a function of how the network responds to these statistics, not changes in the training statistics

Simulation Results Compared to Experimental Data

Illusions of Equal Signs When equal sign is on the right When equal sign is on the left Illusory equal signs

Discussion of equivalence simulation At first: – the model exhibits an ‘add all’ strategy, filling in the blank with the sum of the other numbers presented – and it exhibits illusory perception of the = sign in reproducing a = b + __ equations With additional training, even though problems in which the equal sign is on the right predominate, the model gradually comes to overcome both tendencies, as children do as they gain more and more practice with arithmetic

Limitations and Future Directions The models we’ve used so far: – Use a single parallel settling process, whereas mathematical problem solving clearly can involve a sequence of operations – Use representations of number that don’t fully capture what we know about number intuitions – Lack an interface to explicit propositional statements – Lack an interface to visuospatial representations All of these are important gaps – We have our work cut out for us to incorporate these elements into a more complete model of how we acquire mathematical abilities.

Implications for Education Learning robust automatic encoding skills that translate inputs to their meanings takes time and progresses slowly Thus, we cannot expect to achieve expertise overnight Perhaps most importantly, we cannot blame ourselves or the teacher if we do not understand! – Understanding emerges slowly and requires immersion and engagement Teaching should emphasize – Objects and relations in the world that the expressions map onto – Mapping into this world rather than blindly manipulating symbols – Establishing solid ground before building more on top of it – Realizing that things will not seem clear at first but meaning will emerge with practice 47 5?

Muchas Gracias!