`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`LIBERTY MUTUAL INSURANCE CO.
`
`Petitioner
`
`V.
`
`PROGRESSIVE CASUALTY INSURANCE CO.
`
`Patent Owner
`
`Case CBM2012-00002 (IL)
`Patent 6,064,970
`
`Declaration of Dr. Mark Ehsani
`
`Progressive Exhibit 2016
`
`Liberty Mutual V. Progressive
`CBM 2012-00002
`
`
`
`Declaration of Dr. Mark Ehsani
`
`I, Dr. Mark Ehsani, hereby declare under penalty of perjury:
`
`Scope of Assignment
`
`1)
`
`I was retained by the law firm of Jones Day, on behalf of the Progressive
`
`Casualty Insurance Company (“Progressive”), to render Opinions regarding
`
`fuzzy logic technology.
`
`2)
`
`All of my statements and opinions herein are based on my training and
`
`education as a Ph.D. in Electrical Engineering and as a Professor of
`
`Electrical Engineering and Director of the Advanced Vehicle Systems
`
`Research Program at Texas A&M, and on my work experience as a
`
`research engineer and as an industry consultant to over 60 domestic and
`
`international companies and agencies.
`
`3)
`
`My retention agreement with Jones Day calls for me to be compensated at
`
`my normal rate of $575 per hour, inclusive of any third party expert service
`
`fees, plus out-of—pocket travel expenses.
`
`Qualifications
`
`4)
`
`I am a Professor of electrical engineering and the founding Director of
`
`Advanced Vehicle Systems Research Program and the Power Electronics
`
`and Motor Drives Laboratory at Texas A&M University. I have worked as
`
`a professor and lecturer at Texas A&M University for 32 years. Prior to
`
`this, from 1974 to 1977, I worked as a research engineer at the Fusion
`
`Research Center, University of Texas, and from 1977 to 1981, I worked as
`
`a resident research associate at the Argonne National Laboratory, Argonne,
`
`Illinois. During my time at the Argonne National Laboratory, I was also
`
`performing doctoral work at the University of Wisconsin-Madison in
`
`
`
`energy systems and control systems. My current research work is in power
`
`electronics, motor drives, vehicle electronics, and hybrid vehicles and their
`
`control systems. I have also performed research work in power electronics
`
`and motor drives with regard to applications such as wind power, space
`
`systems, military systems, power and energy storage, and consumer
`
`products, among others.
`
`I have received grants of over $16,000,000 for
`
`filnded research since 1981. I am a- consultant to over 60 domestic and
`
`international companies and agencies.
`
`5)
`
`I received a Doctorate. of Philosophy in electrical engineering from the
`
`University of Wisconsin-Madison in 1981. Prior to this, I received 8.8.
`
`and M.S. degrees from the University of Texas at Austin in 1973 "and 1974,
`
`respectively.
`
`I was the recipient of the Prize Paper Awards in Static Power
`
`Converters and Motor Drives at the IEEE Industry Applications Society
`
`1985, 1987, and 1992 Annual Meetings. In 1984, I was named the
`
`Outstanding Young Engineer of the Year by the Brazos chapter of the
`
`Texas Society of Professional Engineers.
`
`6)
`
`In 1992, I was named the Halliburton Professor in the College of
`
`Engineering at Texas A&M University. In 1994, I was also named the
`
`Dresser Industries Professor at Texas A&M University. In 2001, I was
`
`selected for the Ruth & William Neely / Dow Chemical Faculty Fellow of
`
`the College of Engineering for 2001-2002, for “contributions to the
`
`Engineering Program at Texas A&M,_ including classroom instruction,
`
`scholarly activities, and professional service.” In 2003, I was selected for
`
`the BP Amoco Faculty Award for Teaching Excellence in the College of
`
`Engineering.
`
`I was also selected for the IEEE Vehicular Society 2001
`
`Avant Garde Award for “Contributions to the theory and design of hybrid
`
`electric vehicles.” In 2003, I was selected for the IEEE Undergraduate
`
`
`
`Teaching Award for “outstanding contributions to advanced curriculum
`
`development and teaching of power electronics and drives.” In 2004, I was
`
`elected to the Robert M. Kennedy endowed Chair in Electrical Engineering
`
`at Texas A&M University. In 2005, I was elected as a Fellow of the
`
`Society of Automotive Engineers (SAE).
`
`7)
`
`I have been a member of the IEEE Power Electronics Society (PELS)
`
`AdCom, past Chairman of the PELS Educational Affairs Committee, past
`
`Chairman of the IEEE-IAS Industrial Power Converter Committee, and
`
`past chairman of the IEEE Myron Zucker Student-Faculty Grant program.
`
`I was the General Chair of the IEEE Power Electronics Specialist
`
`Conference for 1990.
`
`I am the founder of the IEEE Vehicle Power and
`
`Propulsion Conference, the founding chairman of the IEEE Vehicular
`
`Technology Society Vehicle Power and Propulsion Committee, and
`
`chairman of Convergence Fellowship Committees. In 2002, I was elected
`
`to the Board of Governors of the IEEE Vehicular Technology Society.
`
`I
`
`also serve on the editorial board of several technical journals and am the
`
`associate editor of IEEE Transactions on Industrial Electronics and IEEE
`
`Transactions on Vehicular Technology.
`
`8)
`
`I am a Fellow of IEEE, an IEEE Industrial Electronics Society and
`
`Vehicular Technology Society Distinguished Speaker, and an IEEE
`
`Industry Applications Society and Power Engineering Society
`
`Distinguished Lecturer.
`
`I am also a registered professional engineer in the
`
`State of Texas.
`
`9)
`
`I am an author on over 350 publications in pulsed-power supplies, high-
`
`voltage engineering, power electronics, motor drives, and advanced vehicle
`
`systems.
`
`I am a co-author of 16 books on power electronics, motor drives,
`
`and advanced vehicle systems, including “Vehicular Electric Power
`
`
`
`Systems,” Marcel Dekker, Inc., 2003, and “Modern Electric Hybrid
`
`Vehicles and Fuel Cell Vehicles — Fundamentals, Theory, and Design,”
`
`CRC Press, 2004. I have over 30 granted or pending United States and
`
`European Union patents.
`
`10)
`
`In my work, I have used fuzzy logic-type algorithms for a variety of
`
`functions (e.g., to perform real-time vehicle data acquisition, logging, and
`
`analysis for driver-specific drive cycle analysis). Some of my work with
`
`computer and fuzzy logic-type algorithms is reflected in my books and
`
`research publications, including, for example, the following:
`
`0
`
`J. P. Johnson, K. M. Rahman, and M. Ehsani, “Application of a
`
`Clustering Adaptive Fuzzy Logic Controller in a Brushless DC
`
`Drive,” IEEE-IECON’97, New" Orleans, LA, November 1997, pp.
`
`1 001-1005.
`
`I Zhiqiang Xu and Mehrdad Ehsani, “Reconstruction of Effective
`
`Wind Speed for Fixed-Speed Wind Turbines Based On Frequency
`
`Data Fusion,” Canadian Conference in Electrical and Computer
`
`Engineering, Calgary, Canada, September, 2010.
`
`a Short Cycle Time Design of Advanced Motor Drives by the Real
`
`Time Simulation and Hardware in the Loop technologies, a two
`
`day short course offered to the automotive industry in Detroit,
`
`Michigan, Nov. 3-4, 2003.
`
`o M. Ehsani, M. Masten, and I. Panahi, “Stiff System Control: A
`
`New Concept in Real Time Control,” Invited Paper at American
`
`Control Conference, San Diego, CA, May 1999.
`
`I Short Cycle Time Design of Advanced Motor Drives by the Real
`
`Time Simulation and Hardware in the Loop Technologies, a two
`
`
`
`day short course offered to the automotive industry in Detroit,
`
`Michigan, Nov. 3-4, 2003.
`
`E. Havaii, B. F. Yancey, and M. Ehsani “Computer Aided Design
`
`Tool for Electric, Hybrid Electric and Plug-in Hybrid Electric
`
`Vehicles,” IEEE-VPPC 2011, Chicago, 111., Oct. 2011.
`
`K. Butler, K. Stevens, and M. Ehsani, “A Versatile Computer
`
`Simulation Tool for Design and Analysis of Electric and Hybrid
`
`Drive Trains,” SAE Proceedings Electric and Hybrid Vehicle
`
`Design Studies, Book # SP 1243, Paper # 970199, February 1997,
`
`pp. 19—25, Detroit, MI.
`
`M. Ehsani, P. Le Polles, M. S. Arefeen, I. Pitel, and J. D. Van Wyk,
`
`“Computer Aided Design and Application of Integrated LC Filters,”
`
`IEEE Trans. on Power Electronics, Vol. 11, No. 1, January 1996,
`
`pp. 182-190.
`
`K. Butler, K. Stevens, and M. Ehsani, “A Versatile Computer
`
`Simulation Tool for Design and Analysis of Electric and Hybrid
`
`Drive Trains,” SAE Proceedings Electric and Hybrid Vehicle
`
`Design Studies, Book # SP 1243, Paper # 970199, February 1997,
`
`pp. 19—25, Detroit, MI.
`
`K. Butler, M. Ehsani, and P. Kamath, “A Matlab-Based Modeling
`
`and Simulation Package for Electric and Hybrid Electric Vehicle
`
`Design,” Invited Paper for the Special Issue of IEEE Trans. on
`
`Vehicular Technology, Vol. 48, No. 6, Nov. 1999, pp. 1770-1778.
`
`Husain and M. Ehsani, “Error Analysis in Indirect Rotor Position
`
`Sensing of Switched Reluctance Motors,” IEEE Trans. on
`
`Industrial Electronics, Vol. 41, No. 3, June 1994, pp. 301-307.
`
`
`
`I M. S. Arefeen, M. Ehsani, and T. A. Lipo, “An Analysis of the
`
`Accuracy of Indirect Shaft Sensor for Synchronous Reluctance
`
`Motor,” IEEE Trans. on Industry Applications, Vol. 30, No. 5,
`
`September/October 1994, pp. 1202-1209.
`
`a K. Butler, K. Stevens, and M. Ehsani, “A Versatile Computer
`
`Simulation Tool for Design and Analysis of Electric and Hybrid
`
`Drive Trains,” SAE Proceedings Electric and Hybrid Vehicle
`
`Design Studies, Book # SP 1243, Paper # 970199, February 1997,
`
`pp. 19-25, Detroit, MI.
`
`0 S. Gay and M. Ehsani, “Parametric Analysis of Eddy-Current
`
`Brake Performance with a 2D Analytical Model,” submitted. to
`
`IEEE Transactions on Magnetics.
`
`I M. Ehsani, I. Husain, and K. R. Ramani, “An Analysis of the Error
`
`in Indirect Rotor Position Sensing of Switched Reluctance Motors,’
`
`IEEE~IECON’91, Kobe, Japan, October 1991, pp. 295-300.
`
`I M. S. Arefeen, M. Ehsani, and T. A. Lipo, “An Analysis of the
`
`Accuracy of Indirect Shaft Sensor for Synchronous Reluctance
`
`Motor,” IEEE-IAS’93, 1993, pp. 695-700.
`
`0 T. Kim and M. Ehsani, “An Error Analysis of the Sensorless
`
`Position Estimation for BLDC Motors”, 2003 IEEE-IAS (Industry
`
`Applications Society) Conference, vol. 1, pp. 611-617, Oct. 2003.
`
`11) At this time, I have nearly 40 years of continuous professional experience
`
`in the fields of electronics, motor drives, power electronics, control
`
`systems, vehicle electronics, and hybrid electric vehicles, among others. In
`
`forming the opinions expressed in this report I have relied upon my
`
`education and my nearly 40 years of professional experience. Attached as
`
`
`
`Exhibit 2017 is a copy of my curriculum vitae, which sets forth my
`
`experience, qualifications, and publications.
`
`Materials Reviewed
`
`12) In preparing this Declaration, I have considered the following materials
`
`listed:
`
`a.
`
`The ‘970 patent (Ex. 1001).
`
`b. Kosaka (Ex. 1004).
`
`c. Herrod (Ex. 1007).
`
`d.
`
`Excerpts related to Kosaka from Liberty Mutual’s ‘970 Petition
`
`filed in the CBM 2012-00002 proceeding (Paper 1).
`
`e.
`
`Excerpts related to Kosaka from Progressive’s Preliminary
`
`Response filed in the CBM 2012-00002 proceeding (Paper 8).
`
`f.
`
`The Board’s Institution Decision (Paper 10).
`
`g. Declaration of Mr. Scott Andrews (Ex. 1012).
`
`h. Declaration of Ms. Mary O’Neil (Ex. 1009).
`
`i.
`
`Other materials cited in this Declaration.
`
`Fuzzy Logic Background — Fuzzy Sets vs. Classical or Crisp Sets
`
`13) Fuzzy logic provides an approach that emulates the way the human mind
`
`works. A person does not typically View experiences in the world as
`
`entirely black or white (1 or 0), but also different shades of gray. For
`
`
`
`example, different people may view the temperature of a room as too high,
`
`a little high, a little low, etc.
`
`14)
`
`The classical approach would use a single value to characterize any
`
`temperature value (e.g., any room temperature above 75 degrees
`
`Fahrenheit could be assigned the single crisp value of “high”.) In contrast,
`
`fitzzy logic would use multiple different “fuzzy” values to characterize
`
`these temperatures. In other words, fuzzy logic considers the relative
`
`degree of trueness of each state of comfort, such as “comfortable”, “cold”-
`
`or “hot”, for each given temperature in the entire applicable temperature
`
`range, such as 0 to 120 degrees Fahrenheit. The exact degree to which a
`
`temperature can be considered a “hot” temperature is determined by an
`
`empirically derived fimction called its membership fimction. In this way,
`
`fuzzy logic considers an infinite number of values of membership in the
`
`“hot” membership function. For example, at 80 degrees the room is more
`
`“hot” than “comfortable” and not much “cold”, and at 75 degrees more
`
`“comfortable” than “hot” and not much “cold”. Similarly, at 65 degrees
`
`the room is more “cold” than “comfortable” and not much “hot”. These
`
`“more”, “less”, and “not much” assessments are arrived at from numerical
`
`values of the infinite valued membership functions of “hot”,
`
`“comfortable”, and “cold”. The “hot”, “comfortable”, and “cold”
`
`functions are called linguistic variables. These functions tie the vagaries of
`
`human language to the rigorous mathematical and logical methodology
`
`(algorithm) that can be used in a computer.
`
`15)
`
`Additional membership fiinctions are typically constructed to describe
`
`more completely the fuzzy variable of interest. (e. g., room temperature). In
`
`this example, a “healthy” fuzzy logic membership fimction could be
`
`established which would also consider an infinite number of degrees of
`
`
`
`membership. Other membership functions could include an “unhealthy”
`
`membership function (with a continuum of infinite values), a “dangerous”
`
`membership function (with a continuum of infinite values), etc.
`
`16)
`
`How much a particular temperature is considered “hot,” “comfortable,” or
`
`“cold” is determined by the membership amount the temperature has in
`
`each membership function. For example, 70 degrees Fahrenheit might
`
`have a larger membership value in the “hot” membership filnction than in
`
`the “comfortable” membership function and “cold” membership filnction.
`
`Multiple fiizzy values are generated for a given temperature that
`
`correspond to the amount of membership in each membership function and
`
`output variables and represent them with linguistic variables. This allows
`
`the converting or mapping of crisp data to fuzzy linguistic values which
`
`have varying degrees for the same data point or datum.
`
`17)
`
`Deriving membership functions is difficult and requires an intimate
`
`knowledge of the application area as well as the different types of
`
`membership fimctions for different linguistic variables in that application
`
`area. These membership functions could be triangular trapezoidal,
`
`Gaussian, bell-shaped, sigmoidal, S-curve, and many others. Which type is
`
`to be used is dependent on the application area, the degree of accuracy
`
`required, and the knowledge of the designer. For example, the
`
`“comfortable”, “hot”, and “cold” membership functions may have to be
`
`developed from a large amount of data, gathered from many people in
`
`various geographic locations and then analyzed and converted to a
`
`membership function. The required degree of accuracy of this membership
`
`function will determine the type and amount of membership fimction data
`
`needed and its mathematical form. Detailed knowledge is needed of the
`
`application area, in terms of what data is relevant, how to gather and
`
`
`
`correlate the data, what the end use of the fuzzy logic algorithm requires to
`
`function properly, what mathematical methods are used to fit data into a
`
`mathematical membership function.
`
`Libem Mutual’s POSITA Standards
`
`18)
`
`I understand that the Petitioner (Liberty Mutual) has engaged Mr. Scott
`
`Andrews as an expert for this proceeding. In his declaration, Mr. Andrews
`
`provided the following description of a POSITA in paragraph 17 of his
`
`declaration: “In my opinion, a person of ordinary skill in the vehicle
`
`telematics aspects pertinent to the ‘970 patent (apart from the insurance
`
`cost aspects), as of January 1996, would have at least a B.S. degree in
`
`electrical engineering, computer engineering, computer science or the
`
`equivalent thereof and at least one to two years of experience with
`
`telematics systems for vehicles.” (Ex. 1012 at 1117.)
`
`19)
`
`I understand that the Petitioner (Liberty Mutual) has also engaged Mary L.
`
`O’Neil as an expert for this proceeding. In her declaration, Ms. O’Neil
`
`provided the following description of a POSITA in paragraph 6 of her
`
`declaration: “In my opinion, a person of ordinary skill in the aspects of
`
`insurance pricing pertinent to the ‘970 patent (apart from the vehicle
`
`telematics aspects) would be an individual with at least a B.S. in
`
`Mathematics, or equivalent, with at least 5 years of experience in the
`
`insurance industry setting premiums for auto insurance, and, as a
`
`minimum, as an associate or near-associate in the Casualty Actuarial
`
`Society (e.g., passed at least 4-5 actuarial exams, including the
`
`examination on ratemaking).” (Ex. 1009 at 116.)
`
`Teachings of the Kosaka Reference
`
`10
`
`
`
`20)
`
`Kosaka discloses applying fuzzy logic to evaluate risk for use in
`
`determining a change in insurance premiums for moving bodies (vehicles).
`
`With the use of fuzzy logic, Kosaka asserts that the “risk evaluation values
`
`can be expected to be more accurate, because they are not susceptible to
`
`external noise and the like.” (Ex. 1004 at 000009.)
`
`21)
`
`Figure 1 in Kosaka shows an insurance premium determination system.
`
`The system includes a fuzzy logic unit (3) which generates comprehensive
`
`risk evaluation output based on sensor data. Kosaka’s specification for
`
`Figure 1 does not reveal details of how the comprehensive risk evaluation
`
`output is generated. Kosaka, however, does discuss this with respect to
`
`Figure 9.
`
`22)
`
`Figure 9 of Kosaka shows three fuzzy logic units (FlU-I, FIU-II, and FIU—
`
`III). FIU-I generates fuzzy risk evaluation values (S, M, B) for evaluating
`
`the risk “related to the frontward moving body,” and FIU—II generates
`
`fuzzy risk evaluation values (S, M, B) for evaluating the risk “related to
`
`‘self" internal states.” (Id. at 000008.) Both outputs from FIU-I and FIU-
`
`II are disclosed as fuzzy values: “The output of the first fuzzy logic part
`
`62 and the second fuzzy logic part 64 are conducted to a third fuzzy logic
`
`part 65 as fuzzy input values.” (Id.; emphasis added.) More specifically,
`
`the fuzzy S, M, and B values respectively show the partial membership
`
`assigmnent degrees for the S, M, and B memberships. There are three
`
`fuzzy S, M, and B output values from FIU-I, and three fuzzy S, M, and B
`
`output values from FIU—II.
`
`23)
`
`FIU-III receives the fuzzy input values from FIU-I and FIU-II to generate
`
`an overall fuzzy risk evaluation output. This output is likewise fuzzy. For
`
`example, Figure 10(E) shows the membership function which operates as
`
`the “FIU—III output function.” (Caption for Figure 10(E).) The output of
`
`11
`
`
`
`Figure 10(E)’s membership function comprises three fuzzy values to
`
`indicate the degree of membership in each of the fuzzy logic categories
`
`(“3” which presumably means small comprehensive risk, “M” which
`
`presumably means medium comprehensive risk, and “B” which
`
`presumably means big comprehensive risk). Because of the continuum in
`
`each of the membership functions, there is an infinite number of degrees of
`
`membership in each of these fuzzy categories. For example, one situation
`
`might yield: multiple fiizzy values of 0.2 S, 0.4 M, and 0.1 B; and another
`
`situation might yield: multiple fuzzy values of 0.6 S, 0.1 M, and 0.1 B.
`
`Further, different situations yield an infinite possible set of fuzzy values
`
`and permutations.
`
`24) Figure 11 provides rules for mapping input fuzzy values to output fuzzy
`
`values. For example, Figure 11(C) shows the rules for mapping the fuzzy
`
`values from FIU—I and FIU—II to generate a consequent fuzzy value for
`
`FUI-III. According to the rules table, an “S” fuzzy value from FIU—I and a
`
`“B” fuzzy value from FIU—II would map to an “M” fiJzzy value for FIU-III.
`
`25) Kosaka’s specification discloses that the fuzzy risk evaluation values (S,
`
`M, B) are directly used as fiJzzy values by the insurance premium
`
`determination unit and does not provide any fiirther processing details
`
`involving these fuzzy values such as how they are used within the
`
`insurance premium determination unit.
`
`Teachings of the Herrod Reference
`
`26) Herrod discloses measuring driver acceleration data and placing the driver
`
`into a particular behavioural group based on accumulated acceleration data.
`
`A display panel is provided “which indicates to the driver the group to
`
`which he or she has been assigned.” (Ex. 1007 at 000002.) An advice code
`
`12
`
`
`
`associated with the particular behavioural group is also displayed so that
`
`the driver can use the advice code to learn “how to change his or her
`
`driving habits to reduce accident risk.” (Id)
`
`27) The behavioural groups are disclosed in Herrod as discrete, crisp type
`
`categories so that a driver can be provided with a single, relevant advice
`
`code. Because the groups are crisp, a driver is assigned to only one
`
`behavioural group based upon the relevant measured data.
`
`A fuzzy logic approach is beyond either POSITA put forth by Mr. Andrews or Ms.
`
`O’Neil
`
`28) Kosaka’s approach would have been beyond the level of a POSITA as
`
`established by either Mr. Andrews or Ms. O’Neil. A POSITA (under
`
`either Ms. O’Neil’s or Mr. Andrews’ description) would not likely have
`
`training or understanding of fiizzy logic, let alone the interrelationships
`
`between the fuzzy membership functions of Figure 10 and the fuzzy rule
`
`evaluation charts of Figure 11 in Kosaka. The inconsistencies, omissions,
`
`and defects of Kosaka further make it very difficult to understand Kosaka.
`
`For example, there is no disclosure in Kosaka about how the multiple
`
`fuzzy risk values from Figure 10(E) would be generated from the fuzzy
`
`rule evaluation charts of Figure 11(C).
`
`29) The fuzzy logic approach was obscure in 1996 and Would not have
`
`been known by Mr. Andrews’ POSITA. This is shown in the course work
`
`and work experience of Mr. Andrews’ POSITA. The course work for Mr.
`
`Andrews’ POSITA would not include fuzzy logic. Fuzzy logic was not
`
`(and still is not) a required course or part of a required course for Mr.
`
`Andrews’ POSITA. Mr. Andrews’ POSITA would not know how to use
`
`fuzzy logic let alone how to apply it to the insurance industry because Mr.
`
`13
`
`
`
`Andrews’ POSITA would not have encountered such subject matter in the
`
`POSITA’s course work or work experience.
`
`The Kosaka reference is so defective that a POSITA would not know how to use
`
`Kosaka
`
`30) Due to the deficiencies in Kosaka’s disclosure, not only would a POSITA
`
`(as described in either Ms. O’Neil’s or Mr. Andrews’ declarations) not
`
`understand how the fuzzy risk evaluation values are generated, but a fuzzy
`
`logic expert would also have great difficulty, if not experience a complete
`
`failure, in trying to understand how these values are generated based on the
`
`disclosure of Kosaka.
`
`31)
`
`Kosaka is missing significant details in how the fuzzy logic approach used
`
`in Kosaka actually works. For example, Figures 10(A)-10(E) are
`
`membership functions. The correSponding disclosure in Kosaka’s
`
`specification is sparse and provides no additional details other than what is
`
`provided in these figures with respect to the mathematical specifications of
`
`the membership functions. The membership functions provided by Kosaka
`
`are more symbolic than Specific and can be found in conceptual
`
`introductions to basic fuzzy logic and hence they do not provide any
`
`description as to what the actual membership functions are.
`
`32)
`
`As another example, Figures 11(A)—11(C) are rule charts for each of the
`
`three fuzzy logic units. The corresponding disclosure in Kosaka’s
`
`specification is limited and provides no additional details other than what
`
`is provided in the figures. For example, there is no description as to how
`
`the rules in Figure 11(C) for FIU-III are used with respect to the
`
`membership functions of Figure 10.
`
`14
`
`
`
`Kosaka’s Approach would change the flmdarnental operation of an approach that
`uses crisp groups, such as actuarial classes.
`
`33)
`
`I have been asked to assume that an “actuarial class” has the following
`
`characteristics: an actuarial class, or risk class, is a grouping of risks (i.e.,
`
`insureds) with similar risk characteristics.
`
`34)
`
`In View of such characteristics a POSITA in the field of fuzzy logic would
`
`not consider the teachings of Kosaka when using a crisp group, such as an
`
`actuarial class, and thus would not use Kosaka in combination with another
`
`reference such as Herrod. My reasons for this include the following.
`
`a) Fuzzy Logic and crisp approaches represent diametrically opposite
`
`approaches. Assignment to a single actuarial class for a risk category
`
`is a crisp logic approach in that it makes assignments crisply to a
`
`single actuarial class for that risk category. In other words, use of
`
`actuarial classes is a crisp approach and generates only a single crisp
`
`value (i.e., a single assignment) for a particular risk category. Fuzzy
`
`logic, on the other hand, uses multiple, partial values to show degrees
`
`of membership a variable of interest might have for its membership
`
`functions.
`
`With assignment to only one actuarial class for a particular risk
`
`category, an actuarial approach does not have a mechanism to
`
`generate multiple fuzzy values such as the multiple fuzzy risk
`
`evaluation values of Kosaka nor an ability to understand or use them.
`
`It would constitute a fundamental change. in operation for an actuarial
`
`class approach to use the multiple, partial membership assignment
`
`fuzzy risk evaluation values of Kosaka.
`
`b) There would be a loss of information if a crisp logic approach is used
`
`instead of fiizzy logic. The fimdamental reason why a person uses a
`
`15
`
`
`
`fuzzy logic approach (versus a classical crisp approach) is that the
`
`person believes that the problem does not lend itself well to a crisp
`
`division and representing the problem through a crisp approach would
`
`lead to less accurate results. Accordingly, selection of a fuzzy logic
`
`approach over a crisp actuarial class approach reflects not a general
`
`preference but rather a, totally different philosophical approach in
`
`addressing a problem.
`
`For a person who has decided to use fuzzy logic to solve a problem,
`
`the use of crisp logic would make the system less accurate. Kosaka
`
`itself recognizes this as Kosaka mentions that with the use of fuzzy
`
`logic, the “risk evaluation values can be expected to be more accurate,
`
`because they are not susceptible to external noise and the like.”
`
`(Kosaka at 9.) To then try to. use a single, crisp actuarial class
`
`assignment approach with the multiple fuzzy risk evaluation values of
`
`Kosaka is to go against the very reasons why fuzzy logic was used in
`
`the first place by Kosaka. Additionally, fuzzy logic is inherently less
`
`susceptible to external noise than crisp methods- This is due at least
`
`in part to fuzzy logic being able to handle data when it is less than
`
`crisp.
`
`[Continued on next page]
`
`16
`
`
`
`Still further, Kosaka’s multiple fuzzy risk evaluation values in
`
`combination with a crisp actuarial class assignment approach would
`
`significantly lessen the accuracy of Kosaka’s approach as Kosaka
`
`would be limited to a single output value whereas the multiple filzzy
`
`risk evaluation values of Kosaka each provide information that would
`
`be lost if replaced by a single number that would incompletely
`
`characterize the situation.
`
`Date: May 1, 2013
`
`W
`
`Signature:
`
`17
`
`