Interview 2 Flashcards
What is title of project?
Using Naturalistic Viewing Paradigm to Explore the Role of Conceptual Knowledge in Face Recognition’
W
What is conceputal information of a face?
The conceptual information refers to person-specific details such as someone’s name, occupation, what they are as a person and personality traits like whether are they introverted or extroverted
What is perceputal information of a face?
Perceptual information of a face is information such as the arrangement of someone’s facial features of the face, the colour of someone’s eyes, and the shape of the face (i.e., does someone have a round face or sharp jaw?)
- What is the key research question you are addressing?
The main question I am trying to address is how conceptual knowledge helps in familiar face recognition and where in the brain this knowledge is represented
- Tell us about your PhD project: the research question and how you hope to investigate this/What is issue with PhD and what is your proposed methodology to investigate it? - (9)
My PhD project aims to answer the question which is how conceptual knowledge helps in familiar face recognition and where in the brain this knowledge is represented
The conceptual information refers to person-specific details such as someone’s name, occupation, what they are as a person and personality traits like whether are they introverted or extroverted
Most of the literature has focused on the role of perceptual information (i.e., visual information from someone’s face such as the arrangement of their facial features and how round their face is) in face recognition
What I would like to do with this project is investigate the role of conceptual knowledge on face recognition and there have been limited studies on this which has limitations.
Their limitation is that faces are presented in a controlled experimental setting that does not reflect our experiences in real life of how we encounter faces
The stimuli are static and are linked with artificial conceptual knowledge like name and occupation as well and perceptual and conceptual knowledge may not be separately acquired which is assumed in the paradigms as well as facial stimuli are presented in ‘trials’ that are not assumed to be independent of each other.
Naturalistic task-free paradigms can be used to overcome these limitations where participants’ neural responses are recorded in an fMRI scanner while watching an engaging movie/documentary, however, it is hard to separate conceptual from perceptual.
A recent behavioural study using naturalistic viewing paradigm by Noads and Andrews had participants watch the original or scrambled version of an episode of Life of Mars and tested how well they understood the story of the movie immediately and after a delay and performed a face recognition task. Both groups have same perceptual information but acquire different cocneputal knowledge. They found that original group performed better in constructing narrative of movie than scrambled. But interestingly, original and scrambled performed equally on face recognition task assuming they were using perceptual features for face recognition of the actors. However, after a delay of 4-week, the original group performs better than scrambled indicating use of conceptual knowledge in face recognition after delay.
Thus, our study will utilise naturalistic viewing paradigm of Noad and Andrews’s Life on Mars paradigm while participants’ neural responses are measured in fMRI and use MVPA and ISC to identify the brain network for conceptual information representation and how perceptual and conceptual networks in brain interact using dynamic causal modelling.
- What methodology of your proposed PhD project in more detail - (4)
We will use the Life on Mars naturalistic viewing paradigm (see Noad and Andrews, 2023). However, participants will initially view either the Original or Scrambled version of the movie outside the scanner which they are unfamiliar with. And they are tested on narrative using structured questions of key events of movie and write a detailed narrative.
Participants in both the Original and Scrambled groups will have the same perceptual information but will acquire different conceptual knowledge.
The neural response will then be measured in participants from both groups using fMRI. They will view a movie containing excerpts from previously unseen episodes of Life on Mars.
Participants from both groups will then be scanned after a delay of 4 weeks. On this occasion, they will view a new movie containing excerpts from previously unseen episodes of Life on Mars.
- Who developed naturalistic viewing paradigms?
Hasson et al 2004
- How using naturalistic viewing paradigm in your study overcome limitations of previous neuroimagning studies? - (4)
- The facial stimuli are dynamic and not static
- The movie is more engaging and thus might make participants more cooperative
- The perceptual and conceptual information is integrated and not separated
- The conceptual knowledge the participants learn of actors in movie is more meaningful as it goes with the story of the movie
- Why did original group in Noad and Andrews have higher face recognition performance after a delay? - (2)
This is because original group new memories of the identities of the actors are successfully consolidated over delay and their memories of new actors is enhanced as they acquired information of the actors in a coherent context which leads to them having higher face recognition of identities of actors over a longer period.
This is not the case of scrambled group as they watched the movie in scrambled order and their memories of actors they have obtained is not in a coherent context and leads to less stable recognition of them and thus leads to poor performance
- How are we defining region of interests (ROIs) in your study? - (2)
For an MVPA approach, we are not defining regions of interest apriori as we are using search light approach with MVPA as looking at the whole brain to identify brain regions which correctly classify actor’s identity above chance level.
For DCM, focusing on direction of flow of information between core and extended system. Therefore, these regions needed to be defined. To do this, we can pick up coordinates that show in ISC and MVPA analysis. Picking up those areas showing above chance classification.
- If not mentioning core and extended systems in more detail, we can say? - (2)
In core network, brain regions responsible for identifying identity of someone
In extended network, brain regions responsible for retrieval of person-specific knowledge
- How are you using MVPA in your study/whats the aim? - (3)
Using MVPA, we will ask whether the pattern of response in different regions of the brain can predict the identity of a person.
To do this, we will compare the pattern of response when one identity is viewed in one part of the video with the pattern of response from a different part of the video.
The MVPA analysis will be performed using a searchlight approach on the whole brain to identify brain regions which correctly classify identity above chance level immediately after participants watch movie and after a delay.
- What are the 2 ways to do MVPA? - (2)
region based
search light = draw 6mm sphere and run classifer sphere , contains one or more regions
- Whatis the difference between traditional univarate and MVPA?- (4)
most of the fMRI studies for face recognition use univariate method of analysis in which each voxel is considered in isolation of others while its activation in response to task or stimulus is estimated.
It is, therefore, assumed that difference in response between the two stimuli, for example, familiar vs non-familiar faces can be observed at the singe voxel level.
The possibility that information could be encoded jointly by the voxels is ignored in the univariate method.
The multivariate pattern analysis (MVPA) (Haxby et al., 2001) overcomes this limitation by considering spatially distributed pattern of activity across voxels. The MVPA makes distinctions between the stimuli or tasks based on the pattern of activity
- What is MVPA’s classifer? - (3)
A classifier is machine learning algorithm learns to associate patterns of brain activity with specific labels
Classifier is ‘trained’ on portion of dataset and told that certain brain activity correspond to different labels
The classifier is then tested on separate portion of dataset not used on training to classify in appropriate labels and its accuracy is observed
- What are the key hypotheses driving your PhD project? – MVPA- immediate: - (2)
Since immediately after the movie, both groups of participants predominantly use perceptual information for recognition, the hypothesis is that brain areas of the core network will have above chance classification in both groups and that the performance between the two groups will not be significantly different.
In contrast, brain regions from the extended network will not show above chance classification in either group
- What are the key hypotheses driving your PhD project? – MVPA- delay - (2)
Given that after a delay, there is a difference in the role of conceptual knowledge for recognition, the hypothesis is that brain areas of the extended network will have above chance classification in the Original group and that the performance will be significantly higher compared to the Scrambled group.
Since conceptual knowledge can also activate perceptual system, there will be above chance classification in the core network for Original group and this and this will not be so in the Scrambled group.
- What is inter-subject correlation? - (2)
inter-subject correlation (ISC) as a measure to calculate how consistent the activation of a given brain area across different participants is when they watch the same movie.
Mathematically, the ISC is simply a correlation coefficient between the time-series of activation extracted from a given voxel/brain area from two participants
- What is difference between inter and intra subject correlation? - (2)
Inter-subject = measures consistency of neural responses across different participants in study
Intra subject = measure consistency of neural responses within same individual across different conditions
- Why is inter-subject correlation used in naturalistic viewing paradigm? - (2)
Since the stimuli presentations in these paradigms are not controlled, a challenge lies in the analysis and interpretation of the data.
Hasson et al (Hasson et al., 2004) used inter-subject correlation (ISC) as a measure to calculate how consistent the activation of a given brain area across different participants is when they watch the same movie.
- How are you using ISC/whats your aim? - (2)
The second aim of the proposal is to use Inter-subject correlation (ISC) to assess the role of conceptual information in face identification
To address this question, the neural response to the movies shown without a delay and with a delay will be compared in participants from the Original and Scrambled groups.
- What are the key hypotheses driving your PhD project? – ISC – delay and immediate: - (4)
Since immediately after the movie, both groups of participants predominantly use perceptual information for recognition, the hypothesis is that brain areas of the core network will have high ISC in both groups and that the performance between the two groups will not be significantly different.
In contrast, brain regions from the extended network may not show high ISC in either group.
Given that after a delay, there is a difference in the role of conceptual knowledge for recognition, the hypothesis is that brain areas of the extended network will have higher ISC in the Original compared to the Scrambled group.
Since conceptual knowledge can also activate perceptual system, there may also be higher ISC in the core network for Original group.
- What is the aim of DCM and how it is used - (2)
The third aim is to understand causal interactions between the core and extended network using dynamic causal modelling (DCM) for fMRI
More specifically, the hypotheses to be tested are (i) the flow of information between core and extended network is influenced by conceptual knowledge and (ii) the flow of information has its directionality from extended to core network
- How to identify causal flow of info in DCM in study for core and extended system? - (3)
To identify causal flow of information between the core and extended network for facial recognition, fMRI data and DCM will be used to estimate three types of models:
Forward model (core network drives the extended network), backward model (extended network drives the core network) and forward-backward model (both core and extended networks drives each other).
These models will be estimated and compared against each other, and the best model will be selected.
- What are the key hypotheses driving your PhD project? – DCM - (2)
Since, conceptual knowledge can activate perceptual system, the hypothesis is that the backward model will be the best model in ‘Original’ group for the data acquired after the delay period when conceptual information is dominant for face recognition.
Since the backward flow of information in the Scrambled group would be absent, the best model for the Scrambled group would be the forward model.
- What is the difference between DCM and functional connectivity (correlational)?
Both DCM and functional connectivity give magnitude of strength of relationship of between different variables however what’s unique to DCM is that it gives the direction of flow of info whether its forward, backward or forward-backward which functional connectivity does not tell you
- Why are we not using Granger causality instead of DCM? - (6)
Both causal connectivity and give direciton of flow of info (regression)
o Granger is not based on neural activity and regression and can be applied to any sort of data
o DCM tries to go at neural level,what you are measuring at fMRI is indirect measurement of neural activity by measuring blood flow
o Start a model where area A and area B have neural activity between them and area A is driving area B
In DCM there is a mathematical equation of neural activity to predicted BOLD signal.
Can compare predicted and actual BOLD and if it matches well the model is true and if not correct.
- Are you using within or between subjects in PhD project?
It is a mixed design between the subject (original and scrambled) and within the subject as both groups watch the movie immediately and after a 4-week delay.
- Why is recognizing faces so important? - (2)
The recognition of familiar faces is fundamental for social interactions.
For example, recognizing whether a face is someone you know or a stranger dictates how you are going to interact with them.
- What are the differences between familiar and unfamiliar faces?
There is a difference between familiar and unfamiliar faces as they require less effort in recognition despite their changes in appearance (Bruce, 1982; Young and Burton, 2019) , are detected faster (Gobbini et al., 2013) are processed more automatically (Jackson and Raymond, 2006) and can hold in working memory more accurately (Jackson and Raymond, 2008).
- Wha are some of the behavioural methodologies showing differences of familiar faces? - Gobbini showing familiar faces detected faster than unfamiliar faces? - (4)
o In an experiment doing continuous flash suppression (CFS) task, participants viewed pairs of rapidly alternating images one containing familiar/unfamiliar face and other a house, one image in each pair rendered invisible through CFS making it unconscious
o Task: Participants reported which image (face or house) they “felt” present, even though they couldn’t consciously see it.
o Finding: Participants were significantly faster at detecting the invisible familiar face compared to the invisible unfamiliar face. This suggests that familiar faces can be processed preconsciously, influencing early stages of visual perception.
o Prioritize detection of personally familiar faces even without conscious awareness
- What is evidence showing perceptual information helps in face recognition?
Studies showing perceptual experience helps recognises faces such as repeated experience of the Jennifer Lawerence’s face from different views, illumination and facial expression produce a robust visual representation that enables participants to recognise her in novel images from new viewpoints improved
- What is evidence showing conceptual info helps in face recognition - (3)
- perceptual learning group = saw each face from multiple angles and lighting conditions but received no additional info,
- conceptual learning group had faces with unique name and occupation,
- participants took a face recognition with novel images of familiar faces , participants in conceptual performed better than perceptual learning and control in learning faces from new viewpoints
- What did Haxby and colleagues propose?
They propose a distributed model of face processing that divides brain areas into core and extended network
- What is core network? - (2)
The core network includes includes brain areas such as Fusiform Face Area (FFA), Superior Temporal Sulcus (STS) and occipital face area (OFA), is thought to extract visual features about the face.
Within this network, the STS extracts dynamic features of faces (O’Toole et al., 2002) which change with time (e.g., speech, shifting gaze, expressing emotional expressions), whereas the FFA is involved in extraction of facial features which are invariant (e.g. identity).
- What is extended network?
The role of extended network is to extract ‘higher level’ conceptual information, such as traits and biographical information, associated with the face and includes brain areas such as the anterior temporal lobe (ATL), cingulate and precuneus.
What is role of ATL, cingulate and prenuceus? - (3)
ATL = It retrieves sematnci memories related to people we know such as recalling its name, occupation, past experiences etc..
Cingulate involves in recalling emotions associated with person’s face like feelings of trust/empathy
Precuneus: creates mental representations of situations and past experiences
- What are the practical applications of your PhD project?/Impact of the project - (4)
- The ability to recognize faces is fundamental to our social and emotional development.
- Therefore, understanding face recognition in humans is valuable in the context of mental health, as difficulties in recognizing faces can be indicative of certain neurological or psychological conditions.
Therapies can address this as a misunderstanding of faces such as their emotions and challenges in social interactions can lead to conflicts/ symptomsand therapies and interventions address to improve the social functioning of individuals wi eh h mental health
The findings may also have relevance to face recognition technology.
Efficientand accurate face recognition systems can aid in identifying and tracking individuals fo public safety and crime prevention.
- How many participants are you expected to recruit?
Recruit participants from local population and given sensitivity analyses from similar studies, we estimate that we will need 20-25 participants per condition
- What challenges you might face in the project? - (4)
- Firstly, managing my time more effectively and balancing research commitments as well as other responsibilities that I have is crucial as I transition to a more independent researcher role.
- Secondly, participant recruitment, can be a challenge for all experiments including neuroimaging studies and require proactive strategies to encourage participants to participate
- Also processing and analysing large-scale neuroimaging datasets can be computationally intensive and demanding and ensuring I have good data management skills to ensure I label files correctly
- Data analysis may pose a challenge as well since I will have to analyse these large-scale neuroimaging data as it requires advanced computational skills and familiarity of specialised software of neuroimaging software which I am starting to learn about in my second semester of my degree.
- Why do you want to do a PhD in this field of face recognition? - (4)
Undertaking the proposed PhD to explore the role of conceptual knowledge in face
recognition’ is driven by my intellectual, personal, and professional motivations.
Personally, I have had first-hand experience of role conceptual knowledge has in familiar face recognition. As a person who travels to visit their family back in India, I was astonished that I can still recognize the faces of the uncles and aunties, which must be due at some level to my knowledge of them as people.
My intellectual motivation behind how this occurs at the brain level has led me to apply for this PhD program.
Professionally, completing this project aligns with my career aspirations of making a career as an academic researcher in the field of face
recognition. By completing this PhD program, I will have developed a deep understanding of the neural mechanisms behind face recognition as well as master both multi-voxel techniques and task-free naturalistic paradigms which will equip me with valuable skills that are sought out in world-leading research labs in this field.
- Why do you want to do a PhD in general?
Generally, I am motivated to do a PhD since I have always had the belief of continuing to learn throughout my life as I find learning, especially about face recognition quite exciting and pursuing a PhD aligns with those personal beliefs
Furthermore, I want to do a PhD to deepen my research interests into face perception and this PhD helps me to achieve my long-term career aspiration of becoming a neuroscientist in this field as well as this PhD will open doors for me in terms of networking in able to connect with many individuals who has expertise in face recognition.
- What are the sources of background noises in speech-in-noise perception?
The source of these background noises could be interfering speech sounds (e.g., many people speaking at the same time) or environmental sounds (e.g., loud TV, traffic noises, etc..)
- Why is SiN perception challenging?
It is challenging for listeners since the background noise degrades the speech signal and thus requires a greater degree of focus and attention of the speaker
- What does successful SiN perception require?
Successful SiN perception requires listeners to separate target speech from other irrelevant noises
- What are the mechanisms behind SiN perception/figure-ground separation? - (4)
Teki et al., (2013) proposed the temporal coherence model to explain the mechanisms behind figure detection
The model is based on the fact that the input sound in the ear is first divided parallel frequency channels (from low to high-frequency) that will then transmit auditory information to the brain
If the different frequency channels are transmitted related information tto the brain then it is grouped as a figure and if the information across the channels are not related then these are grouped together and perceived as a background
The temporal coherence model can be applied to both SiN perception as well as figure detection in SFG and HRSFG stimuli.
- Why is there an emphasis to develop SiN tests developed? = SiN Difficulty - (3)
Problems with SiN perception is often referred to as SiN difficulty or more informally the cocktail party effect
SiN difficulty is common among hearing impaired individuals and most associated with higher rates of anxiety symptoms, social isolation and generally lower quality of life as compared to general population
Therefore, how to diagnose SiN difficulty has been an important research topic in speech perception
- What tests were conventionally used to diagnose SiN difficulty and what were their limitations? = PTA and SiN
- (2)
Conventionally, the diagnosis is based on a pure-tone audiometry (PTA) test which measures an individual’s hearing threshold level based on response to pure-tone stimuli which only captured peripheral hearing loss
SiN tests used with speech in multi-talker babbler noise and more realistic to sounds we encounter and measures both impairments in brain processing and peripheral hearing loss.
- What is multi-talker babble noise?
Many people talking at the same time
- Explain more about figure-ground, what does SFG consist of? - (2)
The SFG stimulus consists of a set of tones of varying frequencies that change randomly and a specific subset of tones that remains the same over time.
The fixed frequencies are perceived as the auditory target amidst the ‘background’ of random variation of frequencies
- How does SFG measure SiN perception?
Typically, the SFG measures SiN perception by assessing the ability to detect the presence of the figure
- What are the advantages of SFG compared to SiN tests?
The SFG stimulus has an advantage over SiN tests, in terms of being applicable to a wider range of participants coming from different language backgrounds because it lacks linguistic content
- What did prior research (i.e., Holmes and Griffiths 2019) show with the SFG and its correlation with SiN (sentence) tests?
A previous study by Holmes and Griffiths (2019) using a figure discrimination task they created using SFG found SFG thresholds only modestly positively correlated with sentence-in-noise performance thresholds.
- What made you create the HRSFG from previous research? - (4)
The modest correlation between SFG and sentence in noise thresholds implies:
SFG is a valid measure of SiN perception but could be further improved to stimulate speech more closely.
Thus, this research gave us rationale to create new version of SFG called HRSFG which contains dynamic and harmonic features of speech in figure which SFG did not have (i.e., static pattern as figure [fixed freq over time)
- Why did the figure discrimination task (by Holmes and Griffiths 2019) include gaps in figure?
Since natural speech is compromised of speech segments that are separated by short gaps, Holmes and Griffiths (2019) created a new task where participants were asked to discriminate which of the two SFG stimuli contained a gap in the figure elements.
- Explain more about speech containing dynamic structure and example: - (2)
Speech is dynamic as its frequencies change over time as we are not speaking with a monotonous voice like a robot
Also on average male pitch is 150 Hz but it does not mean males always talk in 150 Hz in a monotonous voice but its pitch may go to 125 or 180 Hz but always average to 150 Hz.
- Explain more about speech containing harmonic structure and example: - (4)
Additionally, one of the important acoustic features of speech is that a significant portion of the speech sound has a harmonic structure.
That is the spectrum of speech has frequencies which are multiples of fundamental frequencies and an associated pitch.
For example, if the fundamental frequency is 200 Hz then the second (400 Hz), third (600 Hz), and fourth (800 Hz) harmonics etc. are also present
For example, imagine you are saying the word “hello” and the vibration of your vocal cords at the starting point of the word “h” sound creates a fundamental frequency which is like a starting note of melody. As you begin to say the rest of the word, your voice does not produce that one main pitch but produces higher-pitched tones of harmonics.
What is pitch?
Pitch refers to the perceived frequency of a sound and is sometimes determined by the physical frequency of a sound
- Give an example of pitch:
Pitch allows us to distinguish between high or low notes in music
- What is frequency (Hz)?
Frequency is the number of cycles of a sound wave that occur in one second
- What is the limitations of the SFG stimuli? - (3)
There is a modest correlation between SFG performance thresholds and speech-in-noise tests suggesting It can be improved to be more speech-like
Additionally, the figure in SFG is static (has a fixed frequency) and has pure-tone components that are not harmonically related to each other.
Therefore, for an SFG stimulus to mimic the speech stimulus, it may be required for the frequencies of the figure to be harmonically related.
- How were our HRSFG stimuli developed?
So we had natural speech of male recording in which they spoke sentences consisting of a structure of name, verb, number, adjective, and noun.
We then extracted the pitch contours of that recording (i.e., how does the male’s pitch change over time) using a software called Praat and we incorporated that pitch going up and done in our HRSFG stimulus which was created using MATLAB.
