BRIEFING PRESENTATION

Dissertation PsyD, TCSPPMarch 14, 2024

By Illeanna holmgren

Faces Once Forgotten: Advancing the UCLA PMIT with AI to Reflect the Diversity of Our World

Clinical Implications

Literature Review

Methods

Subtitle here

Problem Statement

Subtitle here

Questions

Subtitle here

Research Questions

Background

Content

Facial Memory

Ch.1 Background

Facial memory is the ability to encode, store, and retrieve memory representations of faces (Rapcsak, 2003).Plays a pivotal role in social cognition and interpersonal interactions (Hancock et al. 2000).Studies have shown that average person is able to recognize between 1,000 and 10,000 faces (Jenkins et al., 2018).

Understanding Facial Memory: A Cornerstone of Social Cognition

+ info

+ info

+ info

Hello Stranger

Chapter 1: Background

Prosopagnosia or "facial blindness" is a neurological disorder characterized by the inability to recognize faces.Ranges in severity.Prosopagnosia can be congenital or acquired due to brain injury or certain neurological conditions. Adverse effects include: - Challenges in personal relationships- Increased feelings of anxiety, guilt, and isolation (Yardley et al., 2008).

Images from TA and Ms. World Set

Prospaganpsia

Chapter 1: Background

Prospaganpsia

Prosopagnosia Research Center (2020).

Undiag nosed due to :

• Developmental type not seen in MRI
• Lack of Awareness

• Stigma and disbelief
• Not commonly assessed for in Neuropsychology batteries
• limitations of current facial memory assessment tools inadequately accounting for diversity ((Billings, 2019).

Several estimates suggest 1 in 50 people are prosopagnosic

Chapter 1: Background

PMIT

PMIT aims to address the lack of cultural fairness in assessments

Images from "Picture" and "Korean" sets

Originally designed to assess the neuropsychological impact of HIV-1, the PMIT was created to provide equitable and unbiased measures of cognitive abilities across diverse populations (Maj et al., 1991). Now in digital format: Computerized Memory Interference Test (CMIT) (Life Sciences Core Laboratories, 2024).

• Includes diverse stimuli sets (emojis, baby animals 9 different langugaes) adapting to contemporary global cultural.

Three facial stimuli sets ("Mixed Faces." "Male Faces" and "Female Faces")

Importance of Facial Memory Tests

Background

Chapter 1: Background

Limitations of CMIT

Importance of Facial Memory Tests

Background

Images from TA and Ms. World Set

Limitations of CMIT Facial Stimuli Sets:

• Images align with conventional beauty standards,
• Lack of gender diveristy
• A significant oversight given that 1.6% of U.S. adults identify as transgender or nonbinary, with this figure rising to 5.1% among those under 30 (Pew Research Center, 2022).
• Variability in lighting, posture, emotional expression, and attire across stimuli sets may introduce confounding variables that impact the assessment's validity and reliability (Duchaine & Nakayama, 2006).

The American Psychological Association (APA) code of ethics has long recognized the importance of cultural sensitivity in psychological assessments, advocating for measures that uphold their validity and reliability across varied demographic groups (American Psychological Association, 2016).

Importance of Facial Memory Tests

Background

Info

Purpose of the Study:

Literature Review

Warrington Memory Test for Faces (RMT-F)

Validity Concerns

Procedure

Limitations

Developed by Warrington in 1984 to assess facial memory. and aimed to distinguish between right and left hemisphere brain injuries.

Chapter 2:

Stimuli

Procedure

Validitiy

Example of a BFRT item. Subject are to elect three photos that depict the indiivudal pictured above (2, 5 , 6).

Facial Memory Tests

Benton Facial Recognition Test (BFRT)

Developed in the 1960s, the Benton Facial Recognition Test (BFRT) was another standardized measure developed f or the assessment of individuals believed to have acquired prosopagnosia (Benton et al. 1983).

Chapter 2: Literature Review

Facial Memory Tests

Chapter 2: Literature Review

Literature Review

Chapter 2

Cambridge Face Memory Test(CFMT)

Developed in the 1960s, the Benton Facial Recognition Test (BFRT) was another standardized measure developed f or the assessment of individuals believed to have acquired prosopagnosia (Benton et al. 1983).

Validity

Limitations

Malysian

Taiwanese

Women

Australian

Chinese

Stimuli

Original

+ info

ORB

Chapter 2: Literature Review

Literature Review

Chapter 2

The phenomenon by which own-race faces are better recognized than faces of another race (Meissner & Brigham, 2001).Present not only in adults but also in children and infants as young as 3 months old (Kelly et al., 2005; Hayden et al., 2007; Kelly et al., 2007). Theoretical Explanations:

• Social-Cognitive Framework: People categorize faces as in-group or out-group, leading to more detailed processing and better memory for in-group faces (Sporer, 2001).
• Contact Hypothesis: Increased and meaningful contact with other-race individuals can reduce ORB by enhancing familiarity and recognition skills (Wright et al., 2003; Fioravanti-bastos et al. 2014 ).

Limitations and Complexities:Some studies suggest that even significant interracial contact does not guarantee improved recognition of other-race faces, indicating complexities in how ORB is affected by social experience (Jackiw et al., 2008; Rhodes et al., 2009).

Own Race Bias (ORB)

Importance of Facial Memory Tests

Background

ORB in Assesments

Chapter 2: Literature Review

Literature Review

Chapter 2

(Kho et al., 2023)

The own-race bias for face recognition in a multiracial society. Wong et al., (2020),

ORB in CFTRhodes, G., et al. (2009).

ORB in CFTRhodes, G., et al. (2009).

To address the ethnic and cultural biases of many original facial memory aassements, cross-cultural versions have been introduced. The development of these versions acknowledges the importance of the other-race effect in face recognition and aims to create a more inclusive measure of face memory ability (McKone et al., 2011).

Own Race Bias in Facial Memory Tests

ORB in Assesments

Chapter 2: Literature Review

Literature Review

Chapter 2

MCFR:" I enjoyed having different cultures

MCFR: “all the sistas [Black women] looked pleasant.”

RMT-targets look “racists,” and “old slave owners.”

RMT-F: They're all white people?"

RMT-F:"Why are there no Black people?”

RMT-F faces: ugly,” “unattractive,” “untrustworthy,” “angry,” “disturbing” and looking like “mugshots.

Multicultural Facial Recognition Test (MCFR) to evaluate face recognition memory in persons with acquired brain injuries and assess ORB (Billing 2019)Has black-and-white photos of faces from four ethnicities (Asian, Hispanic, White, and African). Participants performed best on same-race faces,Partipants ated the MCFR more favorably compared to the RMT-F.

Multicultural Facial Recognition Test (MCFR) (Billing, 2019).

Chapter 2: Literature Review

Cultural Limitations in Facial Memory Assessments

(O’Hearn et al., 2010).

Limitations of RMT-F for Younger Populations:

• Nearly half of cognitively intact young adults scored in the 10th percentile or less on the RMT-F based on age norms, indicating that the test might not be as effective for younger individuals (O’Bryant, Hilsabeck, McCaffrey, & Gouvier, 2003).

• The test-retest reliability for young adults was below acceptable correlations for a clinical test (O’Bryant et al., 2003).

CFMT Performance in Children:

• Children aged 9 to 12 showed signficant weaker performance on the standard CFMT (O’Hearn et al., 2010).

Development of CFMT for Children (CFMT-C):

• The CFMT-C was created to provide a culturally fair assessment for children.
• Features adjustments such as fewer target faces, a simpler two-choice format, longer exposure times, and child-friendly instructions.

Age Bias

Chapter 2: Literature Review

Cultural Limitations in Facial Memory Assessments

(Scherf et al. 2017)

Participants generally show better recognition for faces of their own gender, known as own-gender bias. While OGB affects both genders, it is reported to be less prevalent among men (McKelvie et al., 1993; Shapiro & Penrod, 1986; Wright & Sladden, 2003).Contested Views on OGB:

• A comprehensive meta-analysis by Shapiro & Penrod (1986) found no significant gender differences in face recognition abilities, challenging the notion of pervasive OGB.

Inclusion of Female Faces in Assessments:

• The CFMT was initially male faces only; the Female Cambridge Face Memory Test (F-CFMT+) was developed in 2017 to better assess facial memory in women.

• Scherf et al. (2017) found no significant performance differences between men and women on the CFMT or F-CFMT+, nor was an own-gender bias observed.

Implications for Test Design:

• These findings challenge previous assumptions about gender differences in facial memory and underline the necessity of integrating gender considerations into psychological assessment designs to achieve more accurate and culturally fair outcomes.

Own Gender Bias (OGB)

Developed initially for assessing visual memory in HIV patients, the PMIT was crafted with the explicit intention of serving a globally diverse population.Imagery Used in PMIT:

• tilizes standardized, culturally neutral line drawings of universally identifiable items, like an apple or an airplane.

Methodology for Image Selection:

• Participants named each picture concept, rated the likeness to their mental representation, and assessed visual complexity.
• Drawings were required to be detailed yet realistic, representing the most typical view of the subject (Snodgrass & Vanderwart, 1980).
• This approach helped determine the most universally recognizable and effective representations of the images

Insturmentaion

• images were printed on 3-by-5-inch cards and manually presented to participants, with instructions given verbally.

Chapter 2: Literature Review

WHO UCLA Picture Memory Interference Test

PMIT evolved into a digital version (CMIT) for improved precision and reliability, employing adjustable timings for more accurate measurements (Franco, 2009).

Led by Dr. Gaston Pfluegl and Dr. Enrique López, the CMIT forms a practical component of research methodology and hypothesis-testing studies.UCLA life sciences undergraduate students are able to examine cognitive performance,and exploe hypotheses about memory and reaction times across demographics using T-tests.

The CMIT retains original imagery and includes new stimuli such as baby animals, emojis, numbers, words in multiple languages, and diverse faces.

The Computer Memory Interefernce Test (CMIT)

UCLA Life Sciences

Evolution to Digital

Integration into Educational Programs

A key study by Karamians (2015) assessed the CMIT for Spanish-speaking individuals, confirming its convergent validity with WMS-III Visual Reproduction subtests. This suggests the CMIT is likely to be a valid measure of assesing for visual memory in spanish speakers , as it aligns well with another test designed to assess the same construct.Another study foiund that Iranian Americans whose first language is Farsi (1FI) performed differently on the PMIT compared to monolingual English-speaking Caucasian students (MFI), indicating that first language impacts performance on CMIT performance .(Kianmahd, 2012).

Increased Diversity

Customers: 789

Diversification of Stimuli:

for Diverse Populations

Validation of CMIT

Chapter 2: Literature Review

The Computer Memory Interefernce Test (CMIT)

Chapter 2: Literature Review

Artificial Intelligence in Psychology

(O’Hearn et al., 2010).

Artifical Intelligence in Clinical Practice

• D’Alfonso (2020): Highlighted AI's ability to predict effective therapeutic approaches by analyzing patient data, allowing for tailored interventions that cater to the specific needs of each patient.
• Miner et al. (2022): Showcased how AI, through analyzing therapy session recordings, can identify areas for therapist improvement, such as enhancing patient engagement and asking critical safety questions.
• Wysa Application: An AI-powered conversational app that employs Cognitive Behavioral Therapy (CBT) principles to provide mental well-being support. Leo et al. (2022) found Wysa effective in reducing symptoms of depression and anxiety, along with physical pain

Chapter 2: Literature Review

Application of AI in Research

AI images in psycholgoicial research

Bias in AI Models

Realism and Diversity of AI-Generated Images:

Levi’s will test AI-generated clothing models to ‘increase diversity’

Ai Applied

Literature Review

RQ2: Is there a statistically significant difference in the facial recognition accuracy between CMIT-AI-Faces and the CMIT- Female Faces?H2: AI-generated Faces will lead to higher facial recognition accuracy compared to traditional photographic stimuli in facial memory tests.

RQ3: Is there a statistically significant difference in the facial recognition accuracy between CMIT-AI-Faces and CMIT- Male Faces?H2: AI-generated Faces will lead to higher facial recognition accuracy compared to traditional photographic stimuli in facial memory tests.

RQ1: Is there a statistically significant difference in the facial recognition accuracy between CMIT-AI-Faces and the CMIT- Mixed Faces?H1: AI-generated Faces will lead to higher facial recognition accuracy compared to traditional photographic stimuli in facial memory tests.

Chapter 2: Research Questions

Chapter 3: Methods

About the Study

Quantitative Research Design:

• Utilizes archival data from UCLA's Computer Memory Interference Tests (CMIT).Employs Analysis of Variance
• Employs Analysis of Variance (ANOVA) tests to investigate memory recall accuracy influenced by different stimuli types: AI-generated faces vs. CMIT-Mixed, Female, and Male Faces

Ppulation and Sample:

• Undergraduate students enrolled in the Life Sciences Core Laboratories course at UCLA, participating in the “Undergraduate Research Initiative for Life Sciences 2.”

Scoring

• Automated scoring calculates true positives, true negatives, false positives, and false negatives across each test part.

Part F

Reaction Time Test)

Part A

Part B

Part C

Part D

Part E

Timeline

Questionaire

Pre-Test.

Book 1.

book 4

Book 2-3.

and hypothesis

Research Questions

Chapter 3: Methods

AI Stimuli

and hypothesis

Research Questions

not based on Us census

+ info

42 male, 42 emale, 42 non binary

Gender

+ info

Ethnic Breakdown

+ info

Young Adults: 18-29 years- 42 imagesMiddle Adults: 30-44 years 42 imagesOlder adults: 45-64 years- 42 images

+ info

127 images

Ages

Number of Images

Stimuli

Methods

Chapter 3: Methods

Creation of Stimuli

and hypothesis

Research Questions

1, Image Creation: SoftwareGetting Ai Prompt: “Generate an image of a neutral-expression face of a 25-year-old female individual of Chinese descent. The face should represent an average appearance, not adhere to commercial standards of attractiveness, and must be consistent with standard lighting and frontal orientation. Showcase unique facial features and hairstyles, that emphasize diversity.”2. Image Enchancmenent: Remini, an AI photo and video enhancer to improve the photos' quality and realism (Splice Video Editor S.r.l., 2023). 3. Photo Alignment Canva, a graphic design platform, was employed to remove the backgrounds of the images, ensuring a uniform backdrop and eyes centered. .

Importance of Facial Memory Tests

Background

ANOVA Tests: Three separate ANOVA tests will be perfomored: Comparison 1 - Mixed Faces vs. AI-Generated FacesComparison 2 - Female Faces vs. AI-Generated FacesCompariosn 3- Male Faces vs. AI-Generated FacesStatistical Software: SPSS Version 26 will be used for all data analysesOrganization of Data

• Demographic variables will be systematically categorized within SPSS to
• Performance data: Will include True Positives (TP) (arrucacry in previously seen image) and True Negatives (TN) (correctly rejecting images) across Books 1, 2, and 3

Ethical Assurances and Data Security Measures

• The archival data, consisting of de-identified information, will be handled in compliance with the ethical standards established by the Life Sciences Laboratories at UCLA

• Data Security Protocols:
• Secure Storage
• Data Encryption
• Secure Data Transfe

Data Processing

Ch. 3 Methods

+ nf=o

+ info

Questions?

Bianchi et al., 2022

FACEGEN in CFMT

Intranasal inhalation of oxytocin improves face processingin developmental prosopagnosia (Bate eet al., 2012)

RQ1: Is there a statistically significant difference in the facial recognition accuracy between CMIT-AI-Faces and the CMIT- Mixed Faces?H1: AI-generated Faces will lead to higher facial recognition accuracy compared to traditional photographic stimuli in facial memory tests.

Levi’s will test AI-generated clothing models to ‘increase diversity’

Levi’s will test AI-generated clothing models to ‘increase diversity’

Ai Applied

Literature Review

Bias in AI Models: Datasets commonly used to train AI algorithms disproportionately consist of images of White individuals as opposed to those from other racial backgrounds (Miller et al. 2023). AI-generated White faces are often perceived as more 'human-like' than real human photographs (Miller et al. 2023).Studies have shown that AI models reflect racial and gender biases (Luccioni et al., 2023).De-biasing Efforts in AI:Development of sophisticated machine learning models that better recognize and process diverse cultural, racial, and gender characteristics (Nguyen & Chen, 2022).Projects like the Inclusive Images Competition, hosted by Google and others, have encouraged the development of AI models that perform well across a diverse range of demographics,

Ethnic Breakdown

Goes beyond broad categories like "Asian" as used in U.S. Census, adopting finer classifications such as "South Asian" and "East Asian" to accurately capture the diversity within groups (Kaneshiro et al., 2011).Further divides these categories by country or region for enhanced diversity and representation.

Age differences in face memory and face processing between younger and older adults in Taiwan. Chinese Cheng et al., 2016).

TFMT used both male and female faces as stimuli in TFMT,Researchers did not find any evidence to suggest that either genderh had a biased advantage in recognizing faces of different gender.

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.

• Lorem ipsum dolor sit amet.
• Consectetur adipiscing elit.
• Sed do eiusmod tempor incididunt ut.

FaceGen enables the creation of faces with a high degree of realism but does not replicate real individuals' faces. The software uses 3D modeling to produce faces that can vary in age, gender, ethnicity, and other facial features. This method of generating faces offers several advantages for research (Bate et al., 2012). An alternate version of the CFMT also used FaceGen faces, and performance on it was highly correlated with performance on the original CFMT (Wilmer, Germine, Loken, et al., 2010).

Original

Chinese

Australian

Realism and Diversity of AI-Generated Images:

Bray et al. (2022), Korshunov & Marcel (2020), Lu et al. (2023), and Shen et al. (2021) have confirmed that high-quality AI-generated images are often indistinguishable from real ones This quality makes them ideal stimuli for psychological assessments.

FaceGen enables the creation of faces with a high degree of realism but does not replicate real individuals' faces. The software uses 3D modeling to produce faces that can vary in age, gender, ethnicity, and other facial features. This method of generating faces offers several advantages for research (Bate et al., 2012). An alternate version of the CFMT also used FaceGen faces, and performance on it was highly correlated with performance on the original CFMT (Wilmer, Germine, Loken, et al., 2010).

Using AI in Psychological Research

Social Perception: Study used StyleGAN to generate diverse facial expressions for Chinese demographics. Study explored the psychological impact of emotions and aging on social perception and interaction (Han et al., 2023) Race and Attractiveness: AI is used to manipulate images to study racial perceptions and the impact of physical attractiveness without ethical concerns of using real photos (Haut et al., 2021; Eberl et al., 2022).

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.

Lorem ipsum dolor sit

Consectetur adipiscing elit

• Lorem ipsum dolor sit amet.
• Consectetur adipiscing elit.
• Sed do eiusmod tempor incididunt ut.