19 Burst results for "NMA"

"nma" Discussed on Brain Inspired

Brain Inspired

03:08 min | 3 months ago

"nma" Discussed on Brain Inspired

"I don't know if it would be harder if If i kind of didn't do any research have research output for a few years and then going back and i i don't know that a lot of people feel that way i mean i think i think A lot it sounded like you went to grad school and so that Sort of more standard way of i guess like becoming a research professor ida. I know where you have any additional comment to wrap up just a for me. The objective was to be able to do my research. And so i did have academic offers and this offer from microsoft research and other offers and what. I chose the one that would enable me to do the research that i wanted to do. And so i just for anyone who might want to try. Academic the sort of pursuing research outside academia. Just say that there are many different paths working at Industry research does not equate coding. At all In fact we had some Computer science grad students finish their. They say that they wrote barely a hundred lines of code because they use maths to prove things in fact there are people at microsoft research to spend most of their time proving that particular algorithms would converge as opposed to writing any code and then they collaborate with someone who's comfortable coating on they would provide maybe the deep learning part and not every industry research requires to be related to product. Something that jane mentioned as well If you want to connect to products you can but that's like extra you don't need to. It's not necessarily required and jianghua saying i'm also completely focused on research on mentoring and doing exactly what i would doing. Academia other than internships are shorter. The students are better paid. So they're in a better mood maybe but And the only thing that. I guess i would miss a little bit. Is i actually enjoy teaching. And like having the same students for five years at being responsible for them as opposed to having them for short stellar. Internships is a different kind of responsibility for the growth of another. I do sense a kind of inclination. Mysore that so. I think at some point i might either go back to academia or take on more of this kind of advising roles and like you know some some forms of teaching that i do but like maybe more damore pets But other than that if someone doesn't like those parts if that's not something that they like there there's really room for doing very good High level research industry go back to academia as soon as you get that down payment for the house. I think that's wise. Not i was not in a research position at girl. I was like a right but it was. It was totally for production. Yeah that's why. I was asking you to begin with guys. This has been a fun. Meandering down the river bed I think it's a really good way to close out the neuro match Panels here so thanks for being.

jianghua microsoft jane Mysore
"nma" Discussed on Brain Inspired

Brain Inspired

01:47 min | 3 months ago

"nma" Discussed on Brain Inspired

"Are we at the beginning. We're at the beginning or The in jaw were able to actually tell us all the things that don't work as joel pointed out we couldn't even get the institutional review board who often says i guess only as me write proposals correctly to say yes to just having it talk to the patients or the subject in a very controlled manner. So we're very very far away from the united saying that y'all here's the actual language they're or generating no morals and then. I think that you gave an example of learning three works was i. Young disallowed single words. Yeah the blocks words. I guess that's where people were at. The beginning of the n are still struggling agree. Jaw yeah. I completely agree right at the beginning. I think that you know this is Many open questions. And you know. I don't think that it trillion providence is going to solve it I think that you know the the real in. We have new breakthroughs really do need to make in terms of you know trying to be able to actually interface with people in inner manner that is productive. Hi well this is great you you guys are. At the beginning. We collectively are at the end. Actually sierra there with the actually so oh by the way he was tony sean. Who used the cottage industry vernacular for like the small labs just rat. Just wrap things up so thanks guys for the wonderful conversation. I really appreciate it and hope that students get get.

joel united Young tony sean
"nma" Discussed on Brain Inspired

Brain Inspired

05:17 min | 3 months ago

"nma" Discussed on Brain Inspired

"So i think one challenged year is the it's Expect that this model congenital arbitrary distribution so you have to give the motto some knowledge about what our test data is going to look like if it is not from your trinidad or twenty. Twenty two distribution I find it very challenging to find a balance between how much prior knowledge you want. Make big team. How generalize bowl this message. So i have some ideas but i don't have very good satisfactory solutions to this problem will go one more. Nlp and then we'll finish with the nursing. Seventy research has always struggled anyway. But the one thing that. I'm struggling at the moment beside scientists to convince my graduating or the the outgoing folks to consider a career that is out outside fang Bed or the outside google By facebook research of microsoft research. Because i don't know. I feel like too many people too many Brightest minds are just going there at the compensation is great than sell them. But i do feel like oldest Gordon can be used for so many interesting problems in problems. But i feel like the too. Many people are just being sucked into this one of five. The tech firms and marsal marginal impact on the society coming out of is individual converging towards zero so. I'm actually struggling to convince them to see beyond these. Pick promise but it's a it's an uphill battle just because of the tech rally that is never ending you on. The compensation looks just so much larger. This is kind of related. So when you start your post doc with yoshua he gave you four options For what to study and the third one was machine translation. And i don't remember if you struggled to accept that. I don't remember you know whether you that you that you up or whether you really you know thought about it a lot or whatever but but that was back then. And now it's all everything's in l. P. that was back then when he wasn't working right or he wasn't yes. Yeah right okay. Translation wasn't so so. I thought what you're gonna say. Is that you struggle to convince your students to do something besides lp or besides you know language based approach because that's the that's the issue that's true as well yeah. It is implied his obser- because all the tech firms are interest service firms yet. Lp's kind of very obvious immediate area. They want to invest in. And then you and probably because of my students are being recorded by companies with the highest level of compensation which is great. But i want. My students were really really bright to do something beyond what i've done. Or what people in our generation or the previous generations have but it's alco battle. I haven't convinced anyone anything except for one student. Who joined tesla that will serve furthest away from the Is not that too far. No no all right. So that's a.

yoshua Gordon microsoft facebook google tesla
"nma" Discussed on Brain Inspired

Brain Inspired

04:04 min | 3 months ago

"nma" Discussed on Brain Inspired

"One difference thing is out for humanitarian is humid attention. It's more explainable. Like i can tell 'em paying attention to this party. Because i think it's important for my house raising the process whereas for Okay maybe there's disagreement on other people company But from shantanu so it's more about maybe making make up nation better or enable interaction between different parts of the input. So thanks from the expanding the benefit perspective There's this difference. Cemented versus michigan l. l. emphasize. There's sort of two different ways. Attention can be used in psychology segregation. I think you're absolutely right on the explain ability side when we think about the one form of controlled attention right but sort of my broader definition of destroying information out is actually much more similar to the transformers version. You know what i think in that in that sense yes. She was right on the money to call attention to that sense. The there's two different definitions most people refer to the controlled version on the biology on the psychology side guys. We were almost out of time here. So i would. I would love to go around and ask you a terrible question here. what What i wanna know is something that you're working on right now that you're really struggling with that seems Just out of your reach that frustrating you that you can't quite grasp joe you wanna you wanna besides your wife being a neuroscientist do you have something else. Good question I think that large opponent to what sort of me it has continued to be like how to you know take the modern neural systems and make them effective in like a real life setting right the we have all these beautiful generative models and you know because of their toxicity controllability in a no win in an ira be setting would allow me to use your an internal review board where. They're going to tell me whether you know i can do. The patients talk to a generative model. Would allow me to actually do this. But i can't get permission for any help any of the healthcare stuff to use these beautiful generative models and but the thing that i'm struggling with the moment is how to reconcile the fact that we have all these very very powerful models and yet in terms of practical use application. They're really really hard. Because we have all these safety toxicity in explain ability problems that stop them from being able you know actually go that last in a half mile tape very good. Who wants to go next. I'm gonna let you guys jump in. Who feels the most frustrated we did. Did you think that the joel built most frustrated is the result after. Roll a die to adjust. I was feeling his wife scientists pain. You know i go next. So what once. I am working out how to make this Languid standing models to generalize to off distribution that. We know that if we if you train them all on space if the motors work on the look similar to your training. But it's hard to make this model. Generalize to nude it. The stewart from the same task.

shantanu michigan joe motors stewart
"nma" Discussed on Brain Inspired

Brain Inspired

03:47 min | 3 months ago

"nma" Discussed on Brain Inspired

"And you want to accomplish a task and do you want to have the person have fun then all of a sudden this opens up a whole new world of problems. I will also just as a small tangent on talk about something that i think for me. His become interesting ally. And i think it's interesting to large amount of the community. Which is that you know the amount of bits that we you to our transfer models versus the amount of bits. That come into a child are very very different. And to some extent i think one of the large challenges that we have at the moment is understanding these new papers on scaling moala's for language models and drain understand that if you look at the scaling laws for people in scaling laws for language models. They're very far apart. And you know part of that could be architectural. Obviously brad knows a lot more about that than i do. But it's still an interesting challenge also in terms of understanding like what does it require You know to to learn and train these agents with maybe we need to go multi-modal or something but it seems very strange that we need you now. Trillions of words in order to train his mom's on the vision side. It's been interesting that for a long time. You know it was speculated. The reason vision models aren't doing better is because we don't have data sets the size of what human experiences like within the first few years of life but now vision data sets her way bigger than wanted person experiences in some cases in their entire life right in there are still running into the same problems. So we've solved the data set size problem. We're now seeing what what other issues remain at least on the inside and i guess it's though language side as well. Do we know what language is for isn't it. isn't it for getting things from people. Isn't it for advancing our own agendas so that we can mate and etc. I want someone to disagree with me by the way others agree with you. I think language is about You know we go through life. And we build models of the world and language gives you the ability to transmit your model of the world to other people and is crucial for inheritance of ideas. I'm happy to be more cynical than that. I think that. That's the way too optimistic. Angelic vantage point. What do you guys think. So i actually one is the so one. Small sub area that i worked on for a couple of years is cold emerging indication that is where we assume that we have multiple agents are trying to solve problems either collectively oriented be. They compete with each other. And then what we do. Is that the unlike usual multi-agency The simulation we give them a preacher reach high with communication channels. But of course. We don't really designating a protocol defense. They have option to evolve or develop their own communication protocols and then one of the goal there is to see these ages collectively can explore. The environment's better collect more richer set of experiences that can be shared notice to solve the problem or compete better against the others asia's and then the reason why i brought up is set. The in one of those works. We are organizing your workshop on Caja jury solve are two to three years in one hotel. We just randomly but it. You're in a year against..

brad asia
"nma" Discussed on Brain Inspired

Brain Inspired

05:17 min | 3 months ago

"nma" Discussed on Brain Inspired

"This is a terrible question to ask. Realize i can't provide perspectives of my perspective. So the first time. I learned about Honor grad for enjoyment. Back propagation in I start mcgrath's in twenty eleven. And that time i think for conversion maybe people are accepting Next big thing but for feel people are definitely not early. Adopters at uc mild applied. We were still your them. Support of machine as far as margie models. Not you're not works until maybe twenty fifty or twenty sixty and i think as brandon mentioned. I agreed that for the pioneers. Like geria- They Lots of trouble didn't have come up this off at tricks to make our work. But i think by the time of tornado. Twenty fourteen for downstream applications are already some mounted practices for tricks. That can make this model work. But somehow at this time i think. So in retrospect i think that was a misconception because my impression was at that time when i was a student is it's extremely difficult to make his model work but when i tried it for the first time around two hundred fifty it's africa not heart i try to set up hyperloop cars And it just works as an other models so i think that audrey slowdown people's adoption up the network models and other balkars Calm people. Don't think that it's in interminable. Or reflecting and properties of language. They may not work as well. But it turns out that Powerful that this symbolic models. Uc's you'll look you know yes thing to make it easy here. Thank you. And i'll say it i'll say jao jao. What was your experience. Yeah so i think a lot of my experience came from you know doing machine. Learning i guess in another field inserted transferring over to naturally which processing when they started grad school finance. It was all about feature engineering and feature engineering. At the time was like the thing that i knew how to do. In support vector machines griller action..

mcgrath brandon jao jao audrey africa Uc
"nma" Discussed on Brain Inspired

Brain Inspired

05:14 min | 3 months ago

"nma" Discussed on Brain Inspired

"Agents is wonderful agent boosters. Who are you job. what would you do. what are you must powder etcetera. Yeah so. I'm an assistant professor at nyu but in this stern business school in the on technology operations statistics department particularly the technology. Subgroup in yemen allotted. My work focuses on Conversational agents in a lot on evaluation of them. I think i'm probably best known for evaluating conversational agents and doing a lot of work On evaluation and you know my focus. The business school aspect of my focus is really more on understanding how these conversational agents how to use them properly within business and particularly interested in healthcare and public health. So i do a lot of collaboration with people in public health. And so that's probably would best known for Okay i'll answered in two parts The first one is what was the most fun So the most fun project that worked out so i worked on some work early in my on on spectral sign. Clustering you have to describe that breed very briefly. Describe that place. Sure sure so sues these this work on on using taking the similarity between different items and then figuring out how to take that similarity matrix and use the most common features using singular value decomposition to cluster items together. The signed part is figuring out how to incorporate opposing a signal so instead of similarity dissimilarity into this matrix in cluster bombs so that's called the spectrum science spectral clustering sound but the fun part about that so the science spectrum festering is cool but the fun part is one of the biggest problems that i see. All my friends having is figuring out how to assign people to tables in a wedding. And so i worked with a friend of mine on actually figuring out how do change the sign spectral clustering into constrained way in order to figure out the wedding seating problem. Which is you have a whole bunch of people Family and relatives..

nyu yemen
"nma" Discussed on Brain Inspired

Brain Inspired

04:00 min | 3 months ago

"nma" Discussed on Brain Inspired

"Generation And since i started the fact that the job also branch bids also working out out to make our national Understanding models robots nowadays. And i think the word can most known for I feel the citations are probably Racia- some Less common applications including textile transfer and pong generation was that. What does that pong generation. Sorry journey dennard pause. Like oh ponderation. Gotcha yeah okay. Yeah that was a fun project The worker like most was slowly build a dialup agent that accurate collaborative his people and solvay. Some puzzles most of us are this task. Orange systems for they help people to book restaurants. Were boogie tickets. But i thought. I pro- fun because they actually design a model that can raise it about each of the human partner antic initiative to give them information they get to complete that task other and recently also When further on the lines of this shimomoto racia- gervais tried to develop estimation models. That can help you amazon. Right more creative taxed. I guess i've been Always interested in this aspect. Came machines cover. That's humas helped Be more productive so he botts you wanna you just wanna fill the internet with more bots right now. The replacing people but healthy people. And that's what they're supposed to but don't they sell us stuff. I'm not. I'm not trying to argue with you but i think bought has a negative connotation in my mind but i think you have a very optimistic outlook on it. Yeah i think they soon Aspects very possible that Just bought a learning free online. They could generate berry on safer toxic on But i think if we try to be about that can help people so yes. It is tricky to find the right application where the bod cam maximum may help humans improve their productivity a poll. Can i also ask some questions on the way everybody. Please jump in and ask questions. Yes so i had actually one kind of conflicting lewis a message security. Is that the okay. So you work on. The january generation connected laissez. Having fun being creative than song in the you talked about helping. Humans be more productive but isn't like they're having fun and kind of opposite notion yet. The machine actually a helpless doing both or does it have to choose one of them. I want them to choose on. But i'm pretty sure it's easier to optimize. Productivity listed live by optimizing Acted stand. It's conflict thing so they can i say by making palmer productive. We're making mess working more. As opposed to as opposed to having fun you. Aren't they the same thing though. Yeah exactly to help you to compete your house in a more I guess it would help with motivation for sure right. I mean if you if people to enjoy the workplace.

Orange systems gervais amazon berry lewis palmer
"nma" Discussed on Brain Inspired

Brain Inspired

05:06 min | 4 months ago

"nma" Discussed on Brain Inspired

"But it remains to be seen whether you know. Steam engines are much simpler than you know. Networks and solve alpha. Gore's assault took a good one hundred years before a good yours. Amex theory was actually useful. I mean now. It's useful right. I h all undergraduates and chemokine jerry. I've taught thermodynamics. it's a very applied subject right now. In spite of being based on this again very elegant math exactly so it may take time for the science of deep learning the catch up to the jury of learning it could be like decades ride on the things that viscount surprising when i tried pulling predecessor to this course. I took a standard rules. Like how fast should the learning rate d. should you emil slowdown with these square limitations. For example which seems like the right theory and it turns out that there's lots of nice theoretical papers and then you try and actually make up all ullmark assignment and it's very frustrating. Because the darn simple clean homework assignment actually contradicts the very elegant theory that says of course think your favorite scaling is linear in iterations. Is it squared of iterations. Whatever it is whatever it is. It's easy to embarrassingly easy to find counter examples and almost harder to find ones that actually support fury for. I wouldn't fairly simple just state that more precisely than i mean. You can't counteract a correct theory. Because it's of course what i was like. What is it about the structure of the real world problem that you're solving. That violates the assumptions of era rights. And that's where. I think like furious. Have to make a substance made progress. And then the question is like how relevant are your. She's right and and and then you know. That's the cycle between through experiment. Right when you find a situation with three doesn't apply because the data has a different assumption generalizes eerie. I think that's incredibly exciting interplay. And so i spent time at facebook also and i talked to a lot of people industry practitioners. They're at facebook sturge..

viscount Gore emil jerry facebook
"nma" Discussed on Brain Inspired

Brain Inspired

05:38 min | 4 months ago

"nma" Discussed on Brain Inspired

"If you do gradient descent starting from small weights heidari paulo meal. You'll get beneficial over fitting. In the sense of jewel interpret data. You'll go through. Every trading point get zero trading error but in between the training points. You won't wiggle as much as you'd think he would. So that in a nutshell the basic idea if you grady descend from small weights. You won't learn that wigley of function. You'll generalize data points. I think that's the simplest way to explain your. I love it. Yes he does magic. And i would say they visit is optimization Stopping the symptom or does is respected rather than going to complex metrics barricade when i teach applied sense always fit a linear model. I see how does it used for the students to distinguish a process. You should always fit a simple linear model and see how does often. It's incredibly hard to beat. The baseline of decently regularized linear logistic regression. Absolutely what he does as and oddly enough if you're doing a neural net then usually the simple neural nets are not as good as the complex ones with as magic. Gradient settled by the way the last time last episode of anura match panel like this i throughout the word magic and got excoriated by careless. Because real magic now real. But i know what you guys mean so i'm not gonna i'm not gonna push back paul. Did you say that magic is the same as mathematics you miss. The quibbles closed called mathematics sterile. In fact so that was fun is magnetic. hey i'm sterile. And i paid for it a little bit of violence. If you don't have lots of bacteria growing in you your association that means you're magic ball but little over sharing everybody but so so sorry. You start with a small ways. You do the gradient descent. Do we even need to regularize or because the the initial idea regularization or the initial need for it was to not over fit. And but what you're saying is with with a big enough model essentially with enough parameters you'll and if you start with a low initial weights. You're knocking over fit. So in that case. Do we even need to regularize. Yeah so you know sometimes no actually like what this what this three tells you is if you don't recognize it all you still won't wildly over fit you can in principle do better with some regularization and we actually have another paper on a three of hydrogen show regression where we show that. If you don't regularize you don't do that bad. But if you recognize optimally you do better in practice. data sets won't always obey all the theoretical you on your model data. To prove your thrones so in practice you always gotta cross validated to hyper brands including requisition brammertz and you just pick in practice The level of regulation that helps you best performance. Your chest.

wigley paulo grady paul
"nma" Discussed on Brain Inspired

Brain Inspired

03:15 min | 4 months ago

"nma" Discussed on Brain Inspired

"One of them is Scientific computing so so not just the mathematics but the actual implementation of course fastball wingfield. Right like every six month the new the the cool new tool comes out. And you have to keep up but there's some foundations there. I feel like i never. I never quite had the formal training in it. Had just been trying my best to pick it up as i go along and and feel always like a little bit behind Yeah so. I wish i'd taken more courses than and then on the opposite end. I wish i were better trained than kind of the same thing in developmental and evolutionary biology Because it's such a foundational thing about trying to study life and gets you know 'cause like i mentioned earlier. I think life is not a computer. Life is life right on and so knowing the history of life such richness of a of observations that people from those fields. I've only recently begun to appreciate it. Because i happen to sit in integrative biology department. I have all these colleagues paeleontology. Just developmental biologist. I hear them talk all i today. How modern learned about this. Why did i not know about this. I want to do that right. My secret dream. Is that my my my confession as biologists is that. I've never done a single our fieldwork in my entire life. Ideal field work is that i take my laptop to a coffee shop to buy fieldwork and so i've been. I've been hinting at all my colleagues. The next time they go out somewhere to collect specimens something. I don't even care what it is. They need to take me along. You should come and collect hydro with us that with will i would love it. You probably competent to do that. Hopefully this we've covered a lot of ground. Actually there's been a lot of fun. Do you guys have any. Did we that. I leave anything out that you were hoping to talk about or or that you feel might be words of wisdom to those students who are listening and after this conversation might be even feeling even more overwhelmed because their list just got a lot longer about what they need to learn it cetera. I i would. I would like to pass along a piece of advice. That i received from my my undergrad research. Advisor grant johnson. So so at the time. I was having this conversation with him because i was i was considering going to grad school hadn't even applied yet and i was like oh i. Do i really want to do this right. And he told me that that england's life when should strive for for for three things. You should pick something that you're good at something that that's interesting and you should be something that's important for the world and that most people succeed in getting approximately two out of three but if he couldn't find something that's three out of free than that's what you should do so I tried to take his advice seriously. Also tried to pass the device along. So i think it's a really good way of not be overwhelmed right because not everyone has to do. Everything was interesting. You is not what's interesting to everyone else. What you're good at Everyone else's good at and even what you consider to be important for world. That is a matter of taste. What you you like. But i think that's that's been a a piece of vices. Come back to my life over and over it. I just want to pass onto or what are you a two or three good day. I think i'm somewhere between two to three okay..

wingfield grant johnson england
"nma" Discussed on Brain Inspired

Brain Inspired

02:40 min | 4 months ago

"nma" Discussed on Brain Inspired

"It goes back to this thing. Time is a fundamental fundamental piece of of behaviors at neural processing. I do hope it's not antithetical to to durham match academy because one thing that we should all understand is that whatever method where using right now is not the be all and end all now. You can try something you'll get somewhere it'll work but there are going to be other methods that come along that are going to be better so never take your method sets dogma. There there's so many great people out there advancing these tools all the time. Now with the help of amazing statisticians that where we're at a moment in time with respect to to the tools that we're using that will evolve further over time as we figure out better ways to do it and so and i hope that that's the way that people have always taught in school that that you need to know the the math of how to build tools but the tools themselves are constantly moving their analogies to some extent for me right these models bill their representations and their analogies and their metaphors for what. We're trying to turn your standing in biological reality. Sometimes i think back to her. I heard this the the rumored that the most modern technology we have in our in our world their computers. And so it's it's an the brain is kind of like the most inscrutable of all organs. Our bodies so we make us analogy that all the brain Work like computer. And so i often have a lot of friends who especially who haven't studied biology much. Try to pin me down. So what is the hardware of the brain. And what is the software Understand terms of computer analogy. And i'd have to explain to them. Well it's not exactly like that like that doesn't actually apply if you take it too literally. And then and then docked before computers when seaman engines was with the most advanced technology that that we have people had the steam engine analogy of the brains it was all pneumatics and seem pressure in hydraulics going going from these two. That are inside your arena. So i don't know. I don't know what the next one is going to be. Maybe when when humanity comes up with a with a more of the technology computers. That will become. I don't know maybe quantum computing and a lot of people like talking about. I have no opinion on the matter but not a lot of people. But there's non zero people who talked about it but in any case. I'm just wondering right looks i don't i can't predict the future technology and i'm wondering by the time that technology It becomes popular whether we will update our analogy of the brain as well because we still will will not have completely understood.

durham match academy
"nma" Discussed on Brain Inspired

Brain Inspired

03:33 min | 4 months ago

"nma" Discussed on Brain Inspired

"Also right and also thinking things were reading for example without a motor output rate. Like even your internal processes. I talked to my cognitive science colleagues. And sometimes i get mad at me but i'll say it anyway like the fact that we have long term memory the fact that we can. We have our internal agree we make decisions. We can imagine a future plan for the future stuff like that. All of that is because the size of our bodies and the nation of our lives say number of decades on how we like a meter tall lips over a couple of decades in order to function that way in the real world. You have to be able to plan for the future. Where a plankton size. We wouldn't happen. Assistant inflicted don't have sophisticated systems likely hydra size. Well i don't know let's s. Adrian seem to have a lot of memory. We it's been it's been very elusive to find any kind of learning in in hydra. and so they. They control their body that they live in a well. They'd be they deliver their nerve netto role being. I'm sorry low. Piedra wanted their learn more about it. I think part of part of this is why i am. In my life's work we've been drawn more to lack of a better term more tractable systems where we could actually kind of access every hard to some from the sensory inputs to their computations and back out to the motor outputs and then have a notion of what the biophysical feedback is like that entire loop that is that is the thing that i actually do. Everyday relic i take inputs that makes disorders like do some stuff that hyphen impact on the world comes back to like that whole thing right rather than studying each part of it by itself Trying to put it together right and that's really really difficult to over the over the timescale. I don't even minutes hours days right but it is possible for relatively accessible since remoter loop since so..

Piedra Adrian
"nma" Discussed on Brain Inspired

Brain Inspired

05:40 min | 4 months ago

"nma" Discussed on Brain Inspired

"Many insightful ideas buried in one thousand nine hundred fifty s papers mccullough of pets at these guys really had thought through a huge number of things that we We're still rediscovering. So i don't know whether the most efficient way is to read all the literature and look for interesting things whether i think we all do need and young people feel bad about this but i think it's true. We kinda need to discover things for ourselves you know. I think most deeply understand something when you figure it out yourself and then okay. It's often very disappointing to realize that this ideas out there already. It'd be you've been scooped or whatever but that doesn't obviate the value of of realizing it uniquely and coming to it through a path of understanding at the top yourself and so i think the idea of being scooped is just not something we should. We should be so distressed. And you've got gotta of us now have a starting point. That's a very different basis discovering new things that if you just read the line in a paper and and then copied it. I would like to add to that by saying that. Perhaps we should also be talking to people in maybe related fields or unexpectedly related. And you know. This is a very sort of traditional route in science right. You just wet yourself to the science track if you were I was raised in india. That's just how it went. So i'd never had occasion to study. You know philosophy of mind for example cognitive neuroscience but in my recent experience i would say maybe like a year and a half bathing back only a year map. I figured that this whole.

india
"nma" Discussed on Brain Inspired

Brain Inspired

05:45 min | 4 months ago

"nma" Discussed on Brain Inspired

"I have optimism. I don't know if it's based or anything. I'm optimistic as well. I think. Adrian brought up about a development earlier. And i think that's actually really interesting clue for me i am. I like agent. I have no training in actual developmental biology. And i'm also terrified by. But but i have this feeling that is super duper important. Because that's what along. Evolution development is wet differentiates bilateral systems. Physical systems fundamentally and so trying to understand the the brain as it is a fully formed thing is is somewhat of a very incomplete picture because it grew that way and the plan which it grew at that way also grew that way over a much longer timeframe and so those are really important constraints in my mind for how we understand it as a whole. Because you can't have a network of neurons that is incapable of getting to the state. It is when it actually does work in the adult animal And we're ignoring neurodegeneration. As mentioned earlier right which is very important partner ten system as well. So i don't i don't have a solution but i think i'm kind of optimistic that once. We can look at the system not as a fully baked thing that works and does thing. That's world characterized by. Think about how it got there. Maybe maybe that's a kind of sideways sideways. Approach to that question. That may be fruitful. I don't know. I think. I think there's a definitely a couple of our colleagues are doing really really cool work computational neuroscience of developing orissa sims and. I'm i'm really quite hopeful. That more people will go into that field and i mean development is such a hard problem. I mean i'm superman. I mean as i mean you. Both kind of confessed to feeling a little bit of a terror to android like. Oh my god. Everything is changing right. The hardware software. Changing yeah it was hard enough as it is and it's still over such a long timescale that would you even do a right and the task that the that the network is supposed to be doing is also changing while the whole time but i think it's the holy grail problem that the two of you working on so so it's kind of amazing. I mean my attempt at working on something that is wanted to law is looking at learning trajectories So you take a fully fledged adult animal. And what does it look like when it's naively placed on in in the task for the first time and over the process of shaping starts to look a lot like the expert animal so what happens then. Can we look at state space representations or can we look at behavioral changes over these weeks that it takes animals from burma task. And tell me if you know. I don't know a mouse lund in seattle or mouse learned in new york are the those trajectories are different. But the final state of the animal having done the tasker identical built scenarios so Note models that can distinguish these learning trajectories. It's it's a topic that's interesting but it's nowhere as complex. I imagine ask the development problem that that you two are attempting. It'll the question of lining is a really interesting one. That comes back to to pulse questions about the interface between single neurons and networks. I've always been intrigued by served. Throwaway line in the at the end of the blissett. Lomo lt paper. Whether so well. Yeah we've characterized this this synoptic plus zero but but they're all kinds of other ways for neurons to be plastic you can..

Adrian burma seattle new york
"nma" Discussed on Brain Inspired

Brain Inspired

03:46 min | 4 months ago

"nma" Discussed on Brain Inspired

"And i believe that that matter snyder and so of course it's possible to end as chemicals. Where because really beautifully shown it's possible to to show that you can get the firing rates from a particular neuron by a model that that works at a. At a somewhat more abstracted level rate you build and networks has you kind of make up the properties of the network had yet you can train all the parameters to do you know. Reproduce the activity of a single neuron in real life. That's that's not actually what's going on. Individual neurons are transforming their inputs with very particular rules that depend on on the underlying mayan channel distributions and as we learn more about about single neuron cell types. The those rules do vary across different brain areas across across cortex under four different cell types. That has to that has to matter. It is possible to build models that don't care about that because they're very high dimensional right so they can kind of sweep that under under the rug. What i think is important about those roles. One is that we may. It may allow neural systems to do complex computations with fewer units that you have all of this interesting molecular machinery under the hood. And so you ton. Every neuron into a complex computational elements so maybe you know maybe that makes it simpler in in some ways because you you have these these rich elements but the other piece.

snyder
"nma" Discussed on Brain Inspired

Brain Inspired

05:05 min | 4 months ago

"nma" Discussed on Brain Inspired

"And i gave him a project connected with the work. We done analyzing the hudson huxley neuron and had an idea about how he could extend their model to have a multiple timescales of adaptation. He would weigh about two weeks later. Came back and said. Yeah i tried. It doesn't really work. But i saw this great talk as at stanford about this technique i map and i think there's this non linear dimensionality reduction taking. I think could be a really interesting way to look into system so often. Running with that very precedent. Actually because that's exactly what we allow us. A of routinely now to analyze all kinds of non linear data. So that's the kind of way that bet experiences like like this costs are random. Seminars can help help ban students introduce new things into the research program in that case it sounds like it was kind of a horizontal direction. Move right because he was kind of technique but often are these getting dragged into different directions. Helplessly and happily i suppose. Yeah well sure yeah dead ends and whatnot but over the course of research career de asks more and more narrow questions into the nuances or bigger and bigger questions. Or is it just Does it run the gamut in my life differently runs the gamut in an example of of a new new direction that a student brought me into she. She was very interesting development and i'm terrified of developments. Everything's changed we love change. I put his a lot of change. And i really. I hadn't ever done any any real reading into that. But she was fascinated by so shaped went off to woods hole. She had been she'd come from cognitive sciences. So didn't really have deep experimental training to enough to neural system behavior or neurobiology at woods hole length how to get into a lab studied new collaboration with someone else to go with bill. Moody in the biology department started working on a whole new system. We had you know three or so papers about that..

stanford Moody
Understanding Your Gut Microbiome with James and Dahlia Marin of Married to Health

No Meat Athlete Radio

02:45 min | 5 months ago

Understanding Your Gut Microbiome with James and Dahlia Marin of Married to Health

"Gut health. And you know the the micro biogas is a little more nuanced than than you guys have taken a little bit. More of a focus There where did that come from. Just pure richest in the science or is that the most effective tool to focus on. I'm going to step off camera because there seems to be some sort of bird reunion happening outside of my window. Seven closed the window. But i'm listening. You say that question or care. So yeah i think the i think honestly was evolution so i just just as we evolved with our eating and started out kind of flexible. -tarian pesca -tarian. We evolved to more hundred whole foods plant based it was. It's the same in our research in our profession muscles registered bags editions. It's like we're constantly digging. I would say we're always looking for that. Where where does this ruin neck about. Why in that. Why connect with other. We're constantly digging. Ultimately we've been digging and it goes to gut health on. I recently claimed in another talk and another project we're working on. We're honestly i think gut health is the nexus of all health for the planet and so it really is these microbes that our foundation where they're talking about soil whether you're talking about our lungs or they're talking about the gut. Where talking about anything in australia. The microbe that makes it exists so we really came that understanding and and as integrative registered dieticians. We are looking for that route. And i think we've found it and one of our goals has always been help. People eat more plants whether they are fully plant based or not how to eat more plans and one of the biggest things is people tells. I won't eat them. I think they taste you know. I like integrating them. I just don't feel well when i eat them and i totally get it. You know coming from that background of not being healthy myself. I didn't even realize that. I had issues until james really was the one who brought it to my attention in. He brought up the fact that i would constantly say i have a stomach. I didn't even realize it at the time. I think it was just kind of natural after eating. I would always have stahnke pre plant based even when we first started plant based and so on my own kind of ideas type journey as while i've understood which foods are triggers for that and working with patients and working with clients over the last check eater or more. We've really started to understand that gut health can really be a hindrance to improving one's whether it's their weight that they wanna lose whether it's their glucose that they want to better manage their diabetes risk factors whether it's their cholesterol cancer risk factors no matter what it is adding more plants as always going to be a great

Australia James Cholesterol Cancer Diabetes
Dr. Scott Harrington on Why a No Oil, Low Fat Diet Is Optimal

No Meat Athlete Radio

02:02 min | 6 months ago

Dr. Scott Harrington on Why a No Oil, Low Fat Diet Is Optimal

"You're in the know oil. Low fat camp i presume alba's understand that and why you arrived that a bad Conclusion yes yes. Okay well this is you know. You have the paleo kito folks who are not that scared of fat and you know and then on the other side you know you have this like whole foods loaf No oil low fat folks. So there's problems with fat namely saturated fat and we know. Trans fats are carcinogenic. And in bad and they're coming out of the diet but saturated fat is another problem that increases your cholesterol by decreasing. Ldl receptors in the liver. So you don't get as much elliot pulled out. So we know we know saturated fats bad but what about oil so oil. Is the coca cola hidden sugar in coca cola of the fat world. Okay so what. I mean by that. Is it's hidden calories and so when you go to the restaurant and you order fajitas for instance you say if he as healthy i'm just gonna order veggie fajitas and it's going to be on a flower and i'm gonna eat it and it's going to be super healthy but you get it and it's dripping oil. Which is i forget. Maybe it's like one hundred and twenty cows per teaspoon tablespoon of oil. And so you're just going to get extra calories now all of oil and vegetable oils When you compare them against animal. Fats are studies. Show that they're they're healthier. They're healthier for you like when it's a head to head comparison you have your lipid profiles in your body are better when when you're when you're eating oil as compared to an animal fat yet but but there again once again so you're gonna usually when people are switching to a plant may side. They're trying to lose weight and get down back to their eighteen year weight and when you hit a plateau you gotta first thing yet check. Is that oil.

Coca Cola Elliot Coca