Journey into EEG
In 2020, during a period of grieving, I became interested in EEG to measure emotions. I began journaling my exploration. My curiosity led me on a long quest for knowledge. The goal– Find out how to give people useful information about their brain– a “Fitbit for the brain.” Here are a handful of journal entries.
First Entry: 1/1/2020: “You can buy a wearable EEG device right now for a few hundred dollars, why hasn’t it changed the world yet?”
1/24/2020: “Innovation can come from: 1) a better way to process the signals in software or 2) with improvements to hardware that make the signals easier to get for everyone. I’m curious where experts think the improvements will come from.”
1/27/2020: “Feeling a bit lost and unsure about my job. I’m interested in solving psychological problems with brain monitoring, but I don’t know if the the non-invasive tech is ready, there are still a bunch of hard problems to solve.”
2/5/2020: “Around 2 months into my exploration, I’ve learned a lot, and have a lot to learn. Problems I think I have- 1) Am still unclear about the possibilities of emotional insights with EEG. I want to know the limitations by the laws of physics of what we can and cannot measure from the scalp. Can we do more than binary choice BCI? Can we correlate brain states along with other measures like eye-tracking or movement-tracking? What are the limitations, seriously? 2) I need experience with real EEG. I’m still unclear about how I could start doing my own experiments.”
3/11/2020: “Met with Tiff Thompson- Her PhD dissertation is the intersection of EEG and depth psychology. I told her I had a dream I was a pioneer of psychoanalytic computational neuroscience. There could be a disconnect between EEG in the clinical (qEEG) and academic settings, there might be secrets here. Left me with the thought- qEEG (Neurofield) is to academic EEG as clinical psych theories (psychoanalysis) is to academic psychology (controlled studies showing how a psych construct operates). Best not to throw out the baby with the bathwater.”
3/20/2020: “We can do better than binary choice, not in controlling software (i.e. brain signals to move left or right), but in measuring brain states. Measures should be selected on 3 criteria: 1) diagnostic (ability of variable to index the target mental state and remain unaffected by related states), 2) sensitivity (ability of variable to respond rapidly to changes in mental state), and 3) reliability (consistency of the neurophysiological inference across individuals and environments). I wonder what variable is best at those factors and tradeoffs.
In BCI applications– the risk is there are many mental and emotional states that may coexist with the one in question and are a source of noise. Brain features can intersect, especially in rapidly changing real time settings. I.e. how do you know you’re measuring just anxiety instead of mental workload or fatigue.
Idea- EEG product that helps long distance couples with emotional insights. (Falls under emotional analytics).”
5/10/2020: “EEG BCI for controlling computers is dead to me.”
7/5/2020: “Call With Scott Cole, PhD in Neuroscience from UCSD: Emotion/cognitive analytics–EEG would add some useful signals. But video processing or how their voice changes is probably just as important if not more important. Individual variation in EEG–Personality? Interesting, wouldn’t be surprised if that’s possible to measure. Maybe intelligence.
Being at the frontier– describing psychological theories with neuroscience, then measuring them in EEG– super powerful.”
8/10/2020: “First day at NeuroField! Excited to get to data collection and analysis. Looking at raw EEG with Dr Dogris, he seems to know things from the squiggles. There might be something here. I am fascinated by this. We’re looking at the EEG analysis of a 25 year old woman who had many concussions. Dr. Dogris was able to see a “mu pattern” and suggests a “frontal lobe disengagement” and a bunch of other things I didn’t understand that he felt were meaningful. Including brain connectivity— what parts of the brain can “talk” to each other. I want to get to the bottom of this.”
8/29/2020: “This whole place (Neurofield) is operating as though there’s a lot of meaning to be unpacked from EEG. There’s that qEEG world and a separate world that thinks EEG is too noisy. I wonder which is correct.
Also, This seems relevant— if decorticate monkeys produce a distinct EEG pattern, then we should be able to differentiate limbic structure from cortical structure. MRI might be able to help us — I could see MRI working together as a sort of calibration for this EEG monitoring.”
10/14/2020: “What if you could measure differences in status/social hierarchy in the brain. High status, high serotonin. Vs. A loser, depression symptoms, shut off, more emotionally reactive and anxious. This could be f***ing huge. This could actually be a match between self-help talk and measurable neuroscience.”
10/17/2020: “When recording an EEG, 50% of the data arriving at the site of the electrode is from neurons directly underneath it, and 95% of the data comes from within 6 centimeters distance (that’s a circle with a diameter of 2 1/3 inches—big). – Duffy (1989).”
12/22/2020: “Idea- Record EEG during a psychotherapy session. Pinpoint emotional responses to topics? Like an ERP matching the audio of the session to brainwaves somehow?”
2/24/2021: “Had a fantastic mentor session with Dr Dogris examining the Independent Components Analysis software, talking through source localization. Scalp EEG captures the sum of locally coherent source dynamics plus non-brain artifacts. There is nothing that suggests to me what I want to do with EEG is impossible. Dr Dogris believes the analysis and machine learning will get even better too. He was showing me how cool source localization actually is. The software takes in all the data and identifies sources in a 3d array, correlating that with brain region. Then he uses UCSD SCCN’s ICLABEL (a classifier program) uses their database of independent components to classify what is going on. SO f***ing cool.”
4/25/2021: “Jaak Panksepp’s work is interesting– 7 basic emotional systems– Seeking, Rage, Fear, Lust, Care, Panic, Play. We are mammals. Can we can find signals for these systems. Panic is related to loneliness, sadness, depression, psychological pain. Attenuated by social bonding chemicals (brain opioids).
I’ve been searching for this sentence “Unsupervised machine learning approaches can discover meaningful structure in data without assigning labels, providing a potentially valuable tool for scientific discovery in mapping biology to psychology. Supervised learning, by contrast, looks for structure in data that matches assigned labels. By comparing the results of supervised and unsupervised machine learning analyses, we can assess the extent to which psychological categories can reasonably be considered the ground truth for what exists in some objective way.”
Lisa Feldman Barret’s work is helping me understand why we haven’t measured biomarkers for emotions– our brains predict the proper response based on the past. Our brain predicts what our actions should be in a given context and we later ascribe the feeling of emotion to those physiological changes. Meaning every emotion is very context and person specific, hence why we can’t find a patterned biomarker. But that seems to be a contradiction to Panksepp’s work.
Maybe you can measure emotions in real time, maybe you can’t, but personality seems more attainable. Stable patterns of cognition.”
5/24/2021: “Can patterns / deviations / individual differences in networks like the default mode network tell us anything interesting about that persons personality or psychology? We seem to be able to measure network activity reasonably well.”
7/9/2021: “Holy sh*t I did a breathwork routine and induced some very strange theta and delta transients–
7/15/2021: “I’m going to investigate this breathwork routine on my friend’s EEGs. Gathering 3 minutes of baseline resting state data and then compare that to active breathwork data.”
7/19/2021: “Currently looking into John Vervaeke’s work on meaning, higher consciousness, psychedelics, transformation, insight. EEG research has focused on “Aha” moments of insight. This breathwork induced altered state of consciousness thing is really cool.”
9/7/2021: “Call with Brad Voytek today–
Does he have any advice for someone with more entrepreneurial interests, is that welcomed at UCSD? He’s starting a data science industry / researcher collaboration thing! Talked about how he was an early employee at Uber and loves talking to investors about ideas and strategy.
Russell Poldrack did an fMRI on himself twice a week for 18 months while measuring other things like cognitive state, cups of coffee, stuff like that. Basically what I’ve been doing with my self-exploration in EEG, but I’m just eyeballing it. Voytek said this is an interesting data science problem worth exploring.
Holy F***, what a conversation. Completely validated my entrepreneurial ambitions and what I’m trying to do with EEG.”
11/15/2021: “Another separate thing Voytek talks about is meaningful patterns of synchronization and desynchronization. Synchronization is related to communication and information organization between brain regions. Too much synchronization is not good, as is too little. There may be disorders of hypersynchronization (i.e., parkinsons, OCD) and disorders of hyposynchronization (schizophrenia).
I believe EEG data is fundamentally important– these dynamics are essential for all kinds of cognitive measures, intelligence. Others– insight. EEG is important in measuring synchrony on tight time scales– neural communication. Insight and intelligence are related to these network dynamics.”
12/1/2021: “Sent in my application to two PhD programs– UCSD Cognitive Science and UCSD Neuroscience.“
12/23/2021: “Emailed Russell Poldrack about his 18 month fMRI study on himself, about possibly doing something similar in EEG– He thinks it should be very interesting, but that EEG might only be sensitive enough to measure global arousal. Even so, this may map onto anxiety. Measuring EEG longitudinally over several months along with self-reported mood, emotional state, diet, heart rate variability, and other measures like anxiety. This is a perfect project to learn neural data science and could be very very cool.”
1/27/2022: “High “entropy” mental states— psychedelic states, early psychosis, divergent and creative thinking. Low entropy— rigid thinking, ocd, depression, anesthesia. Learning how these relate to specific EEG analysis methods like measuring synchrony or complexity will be important.”
2/18/2022: “Well, Dr. Voytek just let me know I didn’t get into either program I applied to. So I asked him what he thinks I should do. He said– apply to work in UCSD labs that use EEG. I also asked him if I could sit in on his class (Neural Oscillations) in the spring. I feel obligated to double down on this path, based on where my interests are naturally directing me.“
4/18/2022: “Was offered a job in an alcohol research lab at UCSD. They are very serious about helping me get into grad school and will let me continue attending Dr. Voytek’s class.“
6/5/2022: “I’ve learned so much about neural time series in Voytek’s class– what the local field potential is and how it’s generated, what the 1/f decay slope means, why you can’t assume that power changes in frequency bands mean that there are any changes to oscillations, and that if you compare brain data to brownian noise you’ll find a stronger than 0 correlation. This all gives me a ton of context for EEG analysis and its interpretation. 1/f slope may correlate with subjective states of worry based on its relation to GABA/Glu balances.“
7/14/2022: “I’ve been learning about the field of alcohol research– predicting future problems with alcohol. Understanding the methods involved in studying a complex, genetically influenced disorder. Our lab’s theoretical predictor is “low response” to alcohol, which seemed to me to contradict other work I’d read about a higher response to alcohol ingestion being the best predictor for future problems. Upon further investigation, they’re both right. They’re different constructs measured at different times in the blood alcohol level response curve. Our lab is measuring response to the depressive/negative effects of alcohol later in the response curve, and the other labs measure response to the stimulant effects of alcohol during a rising blood alcohol response curve. It’s funny how getting to the bottom of perceived contradictions usually ends up giving you a more differentiated view of the field.“
9/12/2022: “Investigating Neal Swerdlow’s work on a behavioral measure related to dopamine function in nucleus accumbens (prepulse inhibition), which is impaired in schizophrenia, ocd, and huntington’s. I wonder if there are any measures like this related to depression/anxiety?”
9/24/2022: “Sent an email to Dr. Justin Riddle, who may be the Neal Swerdlow of depression, who talks about the importance of subjective experience and specific symptoms in relation to biomarkers for mental disorders, rather than treating the disorder like a trait. So he discusses methods for dimensionality discovery– using factor analysis on clinical assessments to determine the structure of the underlying experience of the disorder, and examining their links to brain data, neurochemical systems, and other behavioral measures.”
10/25/2022: “What if you had all different types of cognition in relation to the theoretical brain system involved and did some sort of factor analysis on it. Categories of mental activity– emotions, motivations, feelings. Like could you find that there are like 10 types of thoughts or at least 10 brain systems that could make a thought or experience. Related to Panksepp’s work. Basically, it’s possible you can get a flavor of subjective experience. Maybe you can’t get the exact thought, but could approach the category of thought or purpose. This has probably been done. Maybe anatomically constrained MEG would be good to learn.”
4/27/2023: “Welp, I didn’t get into any PhDs again. Time to triple down and do this Master’s in Barcelona.”
9/22/2023: “I really want to better understand Dr. Gustavo Deco’s work. What is meant by complexity, chaos, order, equilibrium, etc. in the context of brain data?”
12/23/2023: “What’s cool is Richard Gao’s timescale knee parameter corresponds to the decay time constant of the autocorrelation function– and you see timescale differences in the hierarchy of brain areas. Sensory areas = lower timescale, association = higher timescale. The timescale changes due to tasks (working memory). And the magnitude of the change for an individual subject correlates with performance.”
1/16/2024: “Some useful things I’m learning from Gus’s lab– 1) Modeling brain state transitions in models to predict the effects of stimulation. 2) (Shannon) Entropy is NOT the same as entropy production. 3) People have done “neurophenomonology” of subjective experience. 4) Complexity is how many non-redundant patterns there are in the data, related to its compressibility.
Could variability of brain state transitions be correlated with personality?”
3/15/2024: “Irreversibility is a measure of the level of asymmetry in information flow.”
4/26/2024: “If non-equilibrium enables flexible behavior, and flexible behavior is accomplished partly by the “lateral” frontoparietal network, associated with changing attention (unattended trials but still correct). Should have higher irreversibility, at least in this network. But I’m not quite seeing this in the data.”
5/10/2024: “Probabilistic state space– Describes spatiotemporal patterns– take time points and assign a description, like state of synchronization, and then you clusterize them as different positions. Then you describe the positions probabilistically– for a given brain state, how much time are you spending in each. And from there you can describe perturbations that would help transition brain state from non-functional to healthy as described in the grouped data.”
7/25/2024: “Interesting paper — “Phantom oscillations in principal component analysis” — can use this as an argument against using PCA on EEG, this aligns with what I’ve seen.”
7/30/2024: “Thesis defense went AMAZINGLY, better than my practice sessions. Really glad I went to Barcelona and learned so much. This was incredibly satisfying.”
8/15/2024: “Gustavo thinks we should go for a paper and suggested Network Neuroscience. I think a resting-state analysis of irreversibility X drug condition would add interpretability.”
9/1/2024: “This is kind of out there… I just read Buzsaki’s Inside Out book, and related it back to Psychoanalytic Computational Neuroscience. Self-organized intrinsic neural activity biases our experience, which in turn bootstraps context and meaning to that activity in a never ending loop from birth until death. I feel like these dynamics are conducive to the phenomenology of mind described by Carl Jung— synchronicity, archetypes, projection, etc.”
9/12/2024: “Just made a github for my thesis project. I really do want to keep exploring analysis methods, what they mean, and how they’re useful!”
To be continued as a PhD student in YOUR lab…