Researcher: Ila Fiete

Study reveals how an anesthesia drug induces unconsciousness

Anne Trafton |

There are many drugs that anesthesiologists can use to induce unconsciousness in patients. Exactly how these drugs cause the brain to lose consciousness has been a longstanding question, but MIT neuroscientists have now answered that question for one commonly used anesthesia drug.

Using a novel technique for analyzing neuron activity, the researchers discovered that the drug propofol induces unconsciousness by disrupting the brain’s normal balance between stability and excitability. The drug causes brain activity to become increasingly unstable, until the brain loses consciousness.

“The brain has to operate on this knife’s edge between excitability and chaos.” – Earl K. Miller

“It’s got to be excitable enough for its neurons to influence one another, but if it gets too excitable, it spins off into chaos. Propofol seems to disrupt the mechanisms that keep the brain in that narrow operating range,” says Earl K. Miller, the Picower Professor of Neuroscience and a member of MIT’s Picower Institute for Learning and Memory.

The new findings, reported today in Neuron, could help researchers develop better tools for monitoring patients as they undergo general anesthesia.

Miller and Ila Fiete, a professor of brain and cognitive sciences, the director of the K. Lisa Yang Integrative Computational Neuroscience Center (ICoN), and a member of MIT’s McGovern Institute for Brain Research, are the senior authors of the new study. MIT graduate student Adam Eisen and MIT postdoc Leo Kozachkov are the lead authors of the paper.

Losing consciousness

Propofol is a drug that binds to GABA receptors in the brain, inhibiting neurons that have those receptors. Other anesthesia drugs act on different types of receptors, and the mechanism for how all of these drugs produce unconsciousness is not fully understood.

Miller, Fiete, and their students hypothesized that propofol, and possibly other anesthesia drugs, interfere with a brain state known as “dynamic stability.” In this state, neurons have enough excitability to respond to new input, but the brain is able to quickly regain control and prevent them from becoming overly excited.

Woman gestures with her hand in front of a glass wall with equations written on it.
Ila Fiete in her lab at the McGovern Institute. Photo: Steph Stevens

Previous studies of how anesthesia drugs affect this balance have found conflicting results: Some suggested that during anesthesia, the brain shifts toward becoming too stable and unresponsive, which leads to loss of consciousness. Others found that the brain becomes too excitable, leading to a chaotic state that results in unconsciousness.

Part of the reason for these conflicting results is that it has been difficult to accurately measure dynamic stability in the brain. Measuring dynamic stability as consciousness is lost would help researchers determine if unconsciousness results from too much stability or too little stability.

In this study, the researchers analyzed electrical recordings made in the brains of animals that received propofol over an hour-long period, during which they gradually lost consciousness. The recordings were made in four areas of the brain that are involved in vision, sound processing, spatial awareness, and executive function.

These recordings covered only a tiny fraction of the brain’s overall activity, so to overcome that, the researchers used a technique called delay embedding. This technique allows researchers to characterize dynamical systems from limited measurements by augmenting each measurement with measurements that were recorded previously.

Using this method, the researchers were able to quantify how the brain responds to sensory inputs, such as sounds, or to spontaneous perturbations of neural activity.

In the normal, awake state, neural activity spikes after any input, then returns to its baseline activity level. However, once propofol dosing began, the brain started taking longer to return to its baseline after these inputs, remaining in an overly excited state. This effect became more and more pronounced until the animals lost consciousness.

This suggests that propofol’s inhibition of neuron activity leads to escalating instability, which causes the brain to lose consciousness, the researchers say.

Better anesthesia control

To see if they could replicate this effect in a computational model, the researchers created a simple neural network. When they increased the inhibition of certain nodes in the network, as propofol does in the brain, network activity became destabilized, similar to the unstable activity the researchers saw in the brains of animals that received propofol.

“We looked at a simple circuit model of interconnected neurons, and when we turned up inhibition in that, we saw a destabilization. So, one of the things we’re suggesting is that an increase in inhibition can generate instability, and that is subsequently tied to loss of consciousness,” Eisen says.

As Fiete explains, “This paradoxical effect, in which boosting inhibition destabilizes the network rather than silencing or stabilizing it, occurs because of disinhibition. When propofol boosts the inhibitory drive, this drive inhibits other inhibitory neurons, and the result is an overall increase in brain activity.”

The researchers suspect that other anesthetic drugs, which act on different types of neurons and receptors, may converge on the same effect through different mechanisms — a possibility that they are now exploring.

If this turns out to be true, it could be helpful to the researchers’ ongoing efforts to develop ways to more precisely control the level of anesthesia that a patient is experiencing. These systems, which Miller is working on with Emery Brown, the Edward Hood Taplin Professor of Medical Engineering at MIT, work by measuring the brain’s dynamics and then adjusting drug dosages accordingly, in real-time.

“If you find common mechanisms at work across different anesthetics, you can make them all safer by tweaking a few knobs, instead of having to develop safety protocols for all the different anesthetics one at a time,” Miller says. “You don’t want a different system for every anesthetic they’re going to use in the operating room. You want one that’ll do it all.”

The researchers also plan to apply their technique for measuring dynamic stability to other brain states, including neuropsychiatric disorders.

“This method is pretty powerful, and I think it’s going to be very exciting to apply it to different brain states, different types of anesthetics, and also other neuropsychiatric conditions like depression and schizophrenia,” Fiete says.

The research was funded by the Office of Naval Research, the National Institute of Mental Health, the National Institute of Neurological Disorders and Stroke, the National Science Foundation Directorate for Computer and Information Science and Engineering, the Simons Center for the Social Brain, the Simons Collaboration on the Global Brain, the JPB Foundation, the McGovern Institute, and the Picower Institute.

Just thinking about a location activates mental maps in the brain

Anne Trafton |

As you travel your usual route to work or the grocery store, your brain engages cognitive maps stored in your hippocampus and entorhinal cortex. These maps store information about paths you have taken and locations you have been to before, so you can navigate whenever you go there.

New research from MIT has found that such mental maps also are created and activated when you merely think about sequences of experiences, in the absence of any physical movement or sensory input. In an animal study, the researchers found that the entorhinal cortex harbors a cognitive map of what animals experience while they use a joystick to browse through a sequence of images. These cognitive maps are then activated when thinking about these sequences, even when the images are not visible.

This is the first study to show the cellular basis of mental simulation and imagination in a nonspatial domain through activation of a cognitive map in the entorhinal cortex.

“These cognitive maps are being recruited to perform mental navigation, without any sensory input or motor output. We are able to see a signature of this map presenting itself as the animal is going through these experiences mentally,” says Mehrdad Jazayeri, an associate professor of brain and cognitive sciences, a member of MIT’s McGovern Institute for Brain Research, and the senior author of the study.

McGovern Institute Research Scientist Sujaya Neupane is the lead author of the paper, which appears today in Nature. Ila Fiete, a professor of brain and cognitive sciences at MIT, a member of MIT’s McGovern Institute for Brain Research, and director of the K. Lisa Yang Integrative Computational Neuroscience Center, is also an author of the paper.

Mental maps

A great deal of work in animal models and humans has shown that representations of physical locations are stored in the hippocampus, a small seahorse-shaped structure, and the nearby entorhinal cortex. These representations are activated whenever an animal moves through a space that it has been in before, just before it traverses the space, or when it is asleep.

“Most prior studies have focused on how these areas reflect the structures and the details of the environment as an animal moves physically through space,” Jazayeri says. “When an animal moves in a room, its sensory experiences are nicely encoded by the activity of neurons in the hippocampus and entorhinal cortex.”

In the new study, Jazayeri and his colleagues wanted to explore whether these cognitive maps are also built and then used during purely mental run-throughs or imagining of movement through nonspatial domains.

To explore that possibility, the researchers trained animals to use a joystick to trace a path through a sequence of images (“landmarks”) spaced at regular temporal intervals. During the training, the animals were shown only a subset of pairs of images but not all the pairs. Once the animals had learned to navigate through the training pairs, the researchers tested if animals could handle the new pairs they had never seen before.

One possibility is that animals do not learn a cognitive map of the sequence, and instead solve the task using a memorization strategy. If so, they would be expected to struggle with the new pairs. Instead, if the animals were to rely on a cognitive map, they should be able to generalize their knowledge to the new pairs.

“The results were unequivocal,” Jazayeri says. “Animals were able to mentally navigate between the new pairs of images from the very first time they were tested. This finding provided strong behavioral evidence for the presence of a cognitive map. But how does the brain establish such a map?”

To address this question, the researchers recorded from single neurons in the entorhinal cortex as the animals performed this task. Neural responses had a striking feature: As the animals used the joystick to navigate between two landmarks, neurons featured distinctive bumps of activity associated with the mental representation of the intervening landmarks.

“The brain goes through these bumps of activity at the expected time when the intervening images would have passed by the animal’s eyes, which they never did,” Jazayeri says. “And the timing between these bumps, critically, was exactly the timing that the animal would have expected to reach each of those, which in this case was 0.65 seconds.”

The researchers also showed that the speed of the mental simulation was related to the animals’ performance on the task: When they were a little late or early in completing the task, their brain activity showed a corresponding change in timing. The researchers also found evidence that the mental representations in the entorhinal cortex don’t encode specific visual features of the images, but rather the ordinal arrangement of the landmarks.

A model of learning

To further explore how these cognitive maps may work, the researchers built a computational model to mimic the brain activity that they found and demonstrate how it could be generated. They used a type of model known as a continuous attractor model, which was originally developed to model how the entorhinal cortex tracks an animal’s position as it moves, based on sensory input.

The researchers customized the model by adding a component that was able to learn the activity patterns generated by sensory input. This model was then able to learn to use those patterns to reconstruct those experiences later, when there was no sensory input.

“The key element that we needed to add is that this system has the capacity to learn bidirectionally by communicating with sensory inputs. Through the associational learning that the model goes through, it will actually recreate those sensory experiences,” Jazayeri says.

The researchers now plan to investigate what happens in the brain if the landmarks are not evenly spaced, or if they’re arranged in a ring. They also hope to record brain activity in the hippocampus and entorhinal cortex as the animals first learn to perform the navigation task.

“Seeing the memory of the structure become crystallized in the mind, and how that leads to the neural activity that emerges, is a really valuable way of asking how learning happens,” Jazayeri says.

The research was funded by the Natural Sciences and Engineering Research Council of Canada, the Québec Research Funds, the National Institutes of Health, and the Paul and Lilah Newton Brain Science Award.

What is consciousness?

by Rubina Veerakone |

In the hit T.V. show “Westworld,” Dolores Abernathy, a golden-tressed belle, lives in the days when Manifest Destiny still echoed in America. She begins to notice unusual stirrings shaking up her quaint western town—and soon discovers that her skin is synthetic, and her mind, metal. She’s a cyborg meant to entertain humans. The key to her autonomy lies in reaching consciousness.

Shows like “Westworld” and other media probe the idea of consciousness, attempting to nail down a definition of the concept. However, though humans have ruminated on consciousness for centuries, we still don’t have a solid definition (even the Merriam-Webster dictionary lists five). One framework suggests that consciousness is any experience, from eating a candy bar to heartbreak. Another argues that it is how certain stimuli influence one’s behavior.

MIT graduate student Adam Eisen.

While some search for a philosophical explanation, MIT graduate student Adam Eisen seeks a scientific one.

Eisen studies consciousness in the labs of Ila Fiete, an associate investigator at the McGovern Institute, and Earl Miller, an investigator at the Picower Institute for Learning and Memory. His work melds seemingly opposite fields, using mathematical models to quantitatively explain, and thereby ground, the loftiness of consciousness.

In the Fiete lab, Eisen leverages computational methods to compare the brain’s electrical signals in an awake, conscious state to those in an unconscious state via anesthesia—which dampens communication between neurons so people feel no pain or become unconscious.

“What’s nice about anesthesia is that we have a reliable way of turning off consciousness,” says Eisen.

“So we’re now able to ask: What’s the fluctuation of electrical activity in a conscious versus unconscious brain? By characterizing how these states vary—with the precision enabled by computational models—we can start to build a better intuition for what underlies consciousness.”

Theories of consciousness

How are scientists thinking about consciousness? Eisen says that there are four major theories circulating in the neuroscience sphere. These theories are outlined below.

Global workspace theory

Consider the placement of your tongue in your mouth. This sensory information is always there, but you only notice the sensation when you make the effort to think about it. How does this happen?

“Global workspace theory seeks to explain how information becomes available to our consciousness,” he says. “This is called access consciousness—the kind that stores information in your mind and makes it available for verbal report. In this view, sensory information is broadcasted to higher-level regions of the brain by a process called ignition.” The theory proposes that widespread jolts of neuronal activity or “spiking” are essential for ignition, like how a few claps can lead to an audience applause. It’s through ignition that we reach consciousness.

Eisen’s research in anesthesia suggests, though, that not just any spiking will do. There needs to be a balance: enough activity to spark ignition, but also enough stability such that the brain doesn’t lose its ability to respond to inputs and produce reliable computations to reach consciousness.

Higher order theories

Let’s say you’re listening to “Here Comes The Sun” by The Beatles. Your brain processes the medley of auditory stimuli; you hear the bouncy guitar, upbeat drums, and George Harrison’s perky vocals. You’re having a musical experience—what it’s like to listen to music. According to higher-order theories, such an experience unlocks consciousness.

“Higher-order theories posit that a conscious mental state involves having higher-order mental representations of stimuli—usually in the higher levels of the brain responsible for cognition—to experience the world,” Eisen says.

Integrated information theory

“Imagine jumping into a lake on a warm summer day. All components of that experience—the feeling of the sun on your skin and the coolness of the water as you submerge—come together to form your ‘phenomenal consciousness,’” Eisen says. If the day was slightly less sunny or the water a fraction warmer, he explains, the experience would be different.

“Integrated information theory suggests that phenomenal consciousness involves an experience that is irreducible, meaning that none of the components of that experience can be separated or altered without changing the experience itself,” he says.

Attention schema theory

Attention schema theory, Eisen explains, says ‘attention’ is the information that we are focused on in the world, while ‘awareness’ is the model we have of our attention. He cites an interesting psychology study to disentangle attention and awareness.

In the study, the researchers showed human subjects a mixed sequence of two numbers and six letters on a computer. The participants were asked to report back what the numbers were. While they were doing this task, faintly detectable dots moved across the screen in the background. The interesting part, Eisen notes, is that people weren’t aware of the dots—that is, they didn’t report that they saw them. But despite saying they didn’t see the dots, people performed worse on the task when the dots were present.

“This suggests that some of the subjects’ attention was allocated towards the dots, limiting their available attention for the actual task,” he says. “In this case, people’s awareness didn’t track their attention. The subjects were not aware of the dots, even though the study shows that the dots did indeed affect their attention.”

The science behind consciousness

Eisen notes that a solid understanding of the neural basis of consciousness has yet to be cemented. However, he and his research team are advancing in this quest. “In our work, we found that brain activity is more ‘unstable’ under anesthesia, meaning that it lacks the ability to recover from disturbances—like distractions or random fluctuations in activity—and regain a normal state,” he says.

He and his fellow researchers believe this is because the unconscious brain can’t reliably engage in computations like the conscious brain does, and sensory information gets lost in the noise. This crucial finding points to how the brain’s stability may be a cornerstone of consciousness.

There’s still more work to do, Eisen says. But eventually, he hopes that this research can help crack the enduring mystery of how consciousness shapes human existence. “There is so much complexity and depth to human experience, emotion, and thought. Through rigorous research, we may one day reveal the machinery that gives us our common humanity.”

A new computational technique could make it easier to engineer useful proteins

Anne Trafton |

To engineer proteins with useful functions, researchers usually begin with a natural protein that has a desirable function, such as emitting fluorescent light, and put it through many rounds of random mutation that eventually generate an optimized version of the protein.

This process has yielded optimized versions of many important proteins, including green fluorescent protein (GFP). However, for other proteins, it has proven difficult to generate an optimized version. MIT researchers have now developed a computational approach that makes it easier to predict mutations that will lead to better proteins, based on a relatively small amount of data.

Using this model, the researchers generated proteins with mutations that were predicted to lead to improved versions of GFP and a protein from adeno-associated virus (AAV), which is used to deliver DNA for gene therapy. They hope it could also be used to develop additional tools for neuroscience research and medical applications.

Woman gestures with her hand in front of a glass wall with equations written on it.
MIT Professor of Brain and Cognitive Sciences Ila Fiete in her lab at the McGovern Institute. Photo: Steph Stevens

“Protein design is a hard problem because the mapping from DNA sequence to protein structure and function is really complex. There might be a great protein 10 changes away in the sequence, but each intermediate change might correspond to a totally nonfunctional protein. It’s like trying to find your way to the river basin in a mountain range, when there are craggy peaks along the way that block your view. The current work tries to make the riverbed easier to find,” says Ila Fiete, a professor of brain and cognitive sciences at MIT, a member of MIT’s McGovern Institute for Brain Research, director of the K. Lisa Yang Integrative Computational Neuroscience Center, and one of the senior authors of the study.

Regina Barzilay, the School of Engineering Distinguished Professor for AI and Health at MIT, and Tommi Jaakkola, the Thomas Siebel Professor of Electrical Engineering and Computer Science at MIT, are also senior authors of an open-access paper on the work, which will be presented at the International Conference on Learning Representations in May. MIT graduate students Andrew Kirjner and Jason Yim are the lead authors of the study. Other authors include Shahar Bracha, an MIT postdoc, and Raman Samusevich, a graduate student at Czech Technical University.

Optimizing proteins

Many naturally occurring proteins have functions that could make them useful for research or medical applications, but they need a little extra engineering to optimize them. In this study, the researchers were originally interested in developing proteins that could be used in living cells as voltage indicators. These proteins, produced by some bacteria and algae, emit fluorescent light when an electric potential is detected. If engineered for use in mammalian cells, such proteins could allow researchers to measure neuron activity without using electrodes.

While decades of research have gone into engineering these proteins to produce a stronger fluorescent signal, on a faster timescale, they haven’t become effective enough for widespread use. Bracha, who works in Edward Boyden’s lab at the McGovern Institute, reached out to Fiete’s lab to see if they could work together on a computational approach that might help speed up the process of optimizing the proteins.

“This work exemplifies the human serendipity that characterizes so much science discovery,” Fiete says.

“This work grew out of the Yang Tan Collective retreat, a scientific meeting of researchers from multiple centers at MIT with distinct missions unified by the shared support of K. Lisa Yang. We learned that some of our interests and tools in modeling how brains learn and optimize could be applied in the totally different domain of protein design, as being practiced in the Boyden lab.”

For any given protein that researchers might want to optimize, there is a nearly infinite number of possible sequences that could generated by swapping in different amino acids at each point within the sequence. With so many possible variants, it is impossible to test all of them experimentally, so researchers have turned to computational modeling to try to predict which ones will work best.

In this study, the researchers set out to overcome those challenges, using data from GFP to develop and test a computational model that could predict better versions of the protein.

They began by training a type of model known as a convolutional neural network (CNN) on experimental data consisting of GFP sequences and their brightness — the feature that they wanted to optimize.

The model was able to create a “fitness landscape” — a three-dimensional map that depicts the fitness of a given protein and how much it differs from the original sequence — based on a relatively small amount of experimental data (from about 1,000 variants of GFP).

These landscapes contain peaks that represent fitter proteins and valleys that represent less fit proteins. Predicting the path that a protein needs to follow to reach the peaks of fitness can be difficult, because often a protein will need to undergo a mutation that makes it less fit before it reaches a nearby peak of higher fitness. To overcome this problem, the researchers used an existing computational technique to “smooth” the fitness landscape.

Once these small bumps in the landscape were smoothed, the researchers retrained the CNN model and found that it was able to reach greater fitness peaks more easily. The model was able to predict optimized GFP sequences that had as many as seven different amino acids from the protein sequence they started with, and the best of these proteins were estimated to be about 2.5 times fitter than the original.

“Once we have this landscape that represents what the model thinks is nearby, we smooth it out and then we retrain the model on the smoother version of the landscape,” Kirjner says. “Now there is a smooth path from your starting point to the top, which the model is now able to reach by iteratively making small improvements. The same is often impossible for unsmoothed landscapes.”

Proof-of-concept

The researchers also showed that this approach worked well in identifying new sequences for the viral capsid of adeno-associated virus (AAV), a viral vector that is commonly used to deliver DNA. In that case, they optimized the capsid for its ability to package a DNA payload.

“We used GFP and AAV as a proof-of-concept to show that this is a method that works on data sets that are very well-characterized, and because of that, it should be applicable to other protein engineering problems,” Bracha says.

The researchers now plan to use this computational technique on data that Bracha has been generating on voltage indicator proteins.

“Dozens of labs having been working on that for two decades, and still there isn’t anything better,” she says. “The hope is that now with generation of a smaller data set, we could train a model in silico and make predictions that could be better than the past two decades of manual testing.”

The research was funded, in part, by the U.S. National Science Foundation, the Machine Learning for Pharmaceutical Discovery and Synthesis consortium, the Abdul Latif Jameel Clinic for Machine Learning in Health, the DTRA Discovery of Medical Countermeasures Against New and Emerging threats program, the DARPA Accelerated Molecular Discovery program, the Sanofi Computational Antibody Design grant, the U.S. Office of Naval Research, the Howard Hughes Medical Institute, the National Institutes of Health, the K. Lisa Yang ICoN Center, and the K. Lisa Yang and Hock E. Tan Center for Molecular Therapeutics at MIT.

School of Science announces 2024 Infinite Expansion Awards

Anne Trafton |

The MIT School of Science has announced nine postdocs and research scientists as recipients of the 2024 Infinite Expansion Award, which highlights extraordinary members of the MIT community.

The following are the 2024 School of Science Infinite Expansion winners:

  • Sarthak Chandra, a research scientist in the Department of Brain and Cognitive Sciences, was nominated by Professor Ila Fiete, who wrote, “He has expanded the research abilities of my group by being a versatile and brilliant scientist, by drawing connections with a different area that he was an expert in from his PhD training, and by being a highly involved and caring mentor.”
  • Michal Fux, a research scientist in the Department of Brain and Cognitive Sciences, was nominated by Professor Pawan Sinha, who wrote, “She is one of those figurative beams of light that not only brilliantly illuminate scientific questions, but also enliven a research team.”
  • Andrew Savinov, a postdoc in the Department of Biology, was nominated by Associate Professor Gene-Wei Li, who wrote, “Andrew is an extraordinarily creative and accomplished biophysicist, as well as an outstanding contributor to the broader MIT community.”
  • Ho Fung Cheng, a postdoc in the Department of Chemistry, was nominated by Professor Jeremiah Johnson, who wrote, “His impact on research and our departmental community during his time at MIT has been outstanding, and I believe that he will be a worldclass teacher and research group leader in his independent career next year.”
  • Gabi Wenzel, a postdoc in the Department of Chemistry, was nominated by Assistant Professor Brett McGuire, who wrote, “In the one year since Gabi joined our team, she has become an indispensable leader, demonstrating exceptional skill, innovation, and dedication in our challenging research environment.”
  • Yu-An Zhang, a postdoc in the Department of Chemistry, was nominated by Professor Alison Wendlandt, who wrote, “He is a creative, deep-thinking scientist and a superb organic chemist. But above all, he is an off-scale mentor and a cherished coworker.”
  • Wouter Van de Pontseele, a senior postdoc in the Laboratory for Nuclear Science, was nominated by Professor Joseph Formaggio, who wrote, “He is a talented scientist with an intense creativity, scholarship, and student mentorship record. In the time he has been with my group, he has led multiple facets of my experimental program and has been a wonderful citizen of the MIT community.”
  • Alexander Shvonski, a lecturer in the Department of Physics, was nominated by Assistant Professor Andrew Vanderburg, who wrote, “… I have been blown away by Alex’s knowledge of education research and best practices, his skills as a teacher and course content designer, and I have been extremely grateful for his assistance.”
  • David Stoppel, a research scientist in The Picower Institute for Learning and Memory, was nominated by Professor Mark Bear and his research group, who wrote, “As impressive as his research achievements might be, David’s most genuine qualification for this award is his incredible commitment to mentorship and the dissemination of knowledge.”

Winners are honored with a monetary award and will be celebrated with family, friends, and nominators at a later date, along with recipients of the Infinite Mile Award.

A mindful McGovern community

by Julie Pryor |

Mindfulness is the practice of maintaining a state of complete awareness of one’s thoughts, emotions, or experiences on a moment-to-moment basis. McGovern researchers have shown that practicing mindfulness reduces anxiety and supports emotional resilience.

In a survey distributed to the McGovern Institute community, 57% of the 74 researchers, faculty, and staff who responded, said that they practice mindfulness as a way to reduce anxiety and stress.

Here are a few of their stories.

Fernanda De La Torre

Portrait of a smiling woman leaning back against a railing.
MIT graduate student Fernanda De La Torre. Photo: Steph Stevens

Fernanda De La Torre is a graduate student in MIT’s Department of Brain and Cognitive Sciences, where she is advised by Josh McDermott.

Originally from Mexico, De La Torre took an unconventional path to her education in the United States, where she completed her undergraduate studies in computer science and math at Kansas State University. In 2019, she came to MIT as a postbaccalaureate student in the lab of Tomaso Poggio where she began working on deep-learning theory, an area of machine learning focused on how artificial neural networks modeled on the brain can learn to recognize patterns and learn.

A recent recipient of the prestigious Paul and Daisy Soros Fellowship for New Americans, De La Torre now studies multisensory integration during speech perception using deep learning models in Josh McDermott’s lab.

What kind of mindfulness do you practice, how often, and why?

Metta meditation is the type of meditation I come back to the most. I practice 2-3 times per week. Sometimes by joining Nikki Mirghafori’s Zoom calls or listening to her and other teachers’ recordings on AudioDharma. I practice because when I observe the patterns of my thoughts, I remember the importance of compassion, including self-compassion. In my experience, I find metta meditation is a wonderful way to cultivate the two: observation and compassion. 

When and why did you start practicing mindfulness?

My first meditation practice was as a first-year post-baccalaureate student here at BCS. Gal Raz (also pictured above) carried a lot of peace and attributed it to meditation; this sparked my curiosity. I started practicing more frequently last summer, after realizing my mental health was not in a good place.

How does mindfulness benefit your research at MIT?

This is hard to answer because I think the benefits of meditation are hard to measure. I find that meditation helps me stay centered and healthy, which can indirectly help the research I do. More directly, some of my initial grad school pursuits were fueled by thoughts during meditation but I ended up feeling that a lot of these concepts are hard to explore using non-philosophical approaches. So I think meditation is mainly a practice that helps my health, my relationships with others, and my relationship with work (this last one I find most challenging and personally unresolved). 

Adam Eisen

MIT graduate student Adam Eisen.

Adam Eisen is a graduate student in MIT’s Department of Brain and Cognitive Sciences, where he is co-advised by Ila Fiete (McGovern Institute) and Earl Miller (Picower Institute).

Eisen completed his undergraduate degree in Applied Mathematics & Computer Engineering at Queen’s University in Toronto, Canada. Prior to joining MIT, Eisen built computer vision algorithms at the solar aerial inspection company Heliolytics and worked on developing machine learning tools to predict disease outcomes from genetics at The Hospital for Sick Children.

Today, in the Fiete and Miller labs, Eisen develops tools for analyzing the flow of neural activity, and applies them to understand changes in neural states (such as from consciousness to anesthetic-induced unconsciousness).

What kind of mindfulness do you practice, how often, and why?

I mostly practice simple sitting meditation centered on awareness of senses and breathing. On a good week, I meditate about 3-5 times. The reason I practice are the benefits to my general experience of living. Whenever I’m in a prolonged period of consistent meditation, I’m shocked by how much more awareness I have about thoughts, feelings and sensations that are arising in my mind throughout the day. I’m also amazed by how much easier it is to watch my mind and body react to the context around me, without slipping into the usual patterns and habits. I also find mindful benefits in doing yoga, running and playing music, but the core is really centered on meditation practice.

When and why did you start practicing mindfulness?

I’ve been interested in mindfulness and meditation since undergrad as a path to investigating the nature of mind and thought – an interest which also led me into my PhD. I started practicing meditation more seriously at the start of the pandemic to get more first hand experience with what I had been learning about. I find meditation is one of those things where knowledge and theory can support the practice, but without the experiential component it’s very hard to really start to build an understanding of the core concepts at play.

How does mindfulness benefit your research at MIT?

Mindfulness has definitely informed the kinds of things I’m interested in studying and the questions I’d like to ask – largely in relation to the nature of conscious awareness and the flow of thoughts. Outside of that, I’d like to think that mindfulness benefits my general well-being and spiritual balance, which enables me to do better research.

 

Sugandha Sharma

Woman clasping hands in a yoga pose, looking directly into the camera.
MIT graduate student Sugandha Sharma. Photo: Steph Stevens

Sugandha (Su) Sharma is a graduate student in MIT’s Department of Brain and Cognitive Sciences (BCS), where she is co-advised by Ila Fiete (McGovern Institute) and Josh Tenenbaum (BCS).

Prior to joining MIT, she studied theoretical neuroscience at the University of Waterloo where she built neural models of context dependent decision making in the prefrontal cortex and spiking neuron models of bayesian inference, based on online learning of priors from life experience.

Today, in the Fiete and Tenenbaum labs, she studies the computational and theoretical principles underlying cognition and intelligence in the human brain.  She is currently exploring the coding principles in the hippocampal circuits implicated in spatial navigation, and their role in cognitive computations like structure learning and relational reasoning.

When did you start practicing mindfulness?

When I first learned to meditate, I was challenged to practice it every day for at least 3 months in a row. I took up the challenge, and by the end of it, the results were profound. My whole perspective towards life changed. It made me more empathetic — I could step in other people’s shoes and be mindful of their situations and feelings;  my focus shifted from myself to the big picture — it made me realize how insignificant my life was on the grand scale of the universe, and how it was worthless to be caught up in small things that I was usually worrying about. It somehow also brought selflessness to me. This experience hooked me to meditation and mindfulness for life!

What kind of mindfulness do you practice and why?

I practice mindfulness because it brings awareness. It helps me to be aware of myself, my thoughts, my actions, and my surroundings at each moment in my life, thus helping me stay in and enjoy the present moment. Awareness is of utmost importance since an aware mind always does the right thing. Imagine that you are angry, in that moment you have lost awareness of yourself. The moment you become aware of yourself; anger goes away. This is why sometimes counting helps to combat anger. If you start counting, that gives you time to think and become aware of yourself and your actions.

Meditating — sitting with my eyes closed and just observing (being aware of) my thoughts — is a yogic technique that helps me clear the noise in my mind and calm it down making it easier for me to be mindful not only while meditating, but also in general after I am done meditating. Over time, the thoughts vanish, and the mind becomes blank (noiseless). For this reason, practicing meditation regularly makes it easier for me to be mindful all the time.

An added advantage of yoga and meditation is that it helps combat stress by relaxing the mind and body. Many people don’t know what to do when they are stressed, but I am grateful to have this toolkit of yoga and meditation to deal with stressful situations in my life. They help me calm my mind in stressful situations and ensure that instead of reacting to a situation, I instead act mindfully and appropriately to make it right.

The brain may learn about the world the same way some computational models do

Anne Trafton |

To make our way through the world, our brain must develop an intuitive understanding of the physical world around us, which we then use to interpret sensory information coming into the brain.

How does the brain develop that intuitive understanding? Many scientists believe that it may use a process similar to what’s known as “self-supervised learning.” This type of machine learning, originally developed as a way to create more efficient models for computer vision, allows computational models to learn about visual scenes based solely on the similarities and differences between them, with no labels or other information.

A pair of studies from researchers at the K. Lisa Yang Integrative Computational Neuroscience (ICoN) Center at MIT offers new evidence supporting this hypothesis. The researchers found that when they trained models known as neural networks using a particular type of self-supervised learning, the resulting models generated activity patterns very similar to those seen in the brains of animals that were performing the same tasks as the models.

The findings suggest that these models are able to learn representations of the physical world that they can use to make accurate predictions about what will happen in that world, and that the mammalian brain may be using the same strategy, the researchers say.

“The theme of our work is that AI designed to help build better robots ends up also being a framework to better understand the brain more generally,” says Aran Nayebi, a postdoc in the ICoN Center. “We can’t say if it’s the whole brain yet, but across scales and disparate brain areas, our results seem to be suggestive of an organizing principle.”

Nayebi is the lead author of one of the studies, co-authored with Rishi Rajalingham, a former MIT postdoc now at Meta Reality Labs, and senior authors Mehrdad Jazayeri, an associate professor of brain and cognitive sciences and a member of the McGovern Institute for Brain Research; and Robert Yang, an assistant professor of brain and cognitive sciences and an associate member of the McGovern Institute. Ila Fiete, director of the ICoN Center, a professor of brain and cognitive sciences, and an associate member of the McGovern Institute, is the senior author of the other study, which was co-led by Mikail Khona, an MIT graduate student, and Rylan Schaeffer, a former senior research associate at MIT.

Both studies will be presented at the 2023 Conference on Neural Information Processing Systems (NeurIPS) in December.

Modeling the physical world

Early models of computer vision mainly relied on supervised learning. Using this approach, models are trained to classify images that are each labeled with a name — cat, car, etc. The resulting models work well, but this type of training requires a great deal of human-labeled data.

To create a more efficient alternative, in recent years researchers have turned to models built through a technique known as contrastive self-supervised learning. This type of learning allows an algorithm to learn to classify objects based on how similar they are to each other, with no external labels provided.

“This is a very powerful method because you can now leverage very large modern data sets, especially videos, and really unlock their potential,” Nayebi says. “A lot of the modern AI that you see now, especially in the last couple years with ChatGPT and GPT-4, is a result of training a self-supervised objective function on a large-scale dataset to obtain a very flexible representation.”

These types of models, also called neural networks, consist of thousands or millions of processing units connected to each other. Each node has connections of varying strengths to other nodes in the network. As the network analyzes huge amounts of data, the strengths of those connections change as the network learns to perform the desired task.

As the model performs a particular task, the activity patterns of different units within the network can be measured. Each unit’s activity can be represented as a firing pattern, similar to the firing patterns of neurons in the brain. Previous work from Nayebi and others has shown that self-supervised models of vision generate activity similar to that seen in the visual processing system of mammalian brains.

In both of the new NeurIPS studies, the researchers set out to explore whether self-supervised computational models of other cognitive functions might also show similarities to the mammalian brain. In the study led by Nayebi, the researchers trained self-supervised models to predict the future state of their environment across hundreds of thousands of naturalistic videos depicting everyday scenarios.

“For the last decade or so, the dominant method to build neural network models in cognitive neuroscience is to train these networks on individual cognitive tasks. But models trained this way rarely generalize to other tasks,” Yang says. “Here we test whether we can build models for some aspect of cognition by first training on naturalistic data using self-supervised learning, then evaluating in lab settings.”

Once the model was trained, the researchers had it generalize to a task they call “Mental-Pong.” This is similar to the video game Pong, where a player moves a paddle to hit a ball traveling across the screen. In the Mental-Pong version, the ball disappears shortly before hitting the paddle, so the player has to estimate its trajectory in order to hit the ball.

The researchers found that the model was able to track the hidden ball’s trajectory with accuracy similar to that of neurons in the mammalian brain, which had been shown in a previous study by Rajalingham and Jazayeri to simulate its trajectory — a cognitive phenomenon known as “mental simulation.” Furthermore, the neural activation patterns seen within the model were similar to those seen in the brains of animals as they played the game — specifically, in a part of the brain called the dorsomedial frontal cortex. No other class of computational model has been able to match the biological data as closely as this one, the researchers say.

“There are many efforts in the machine learning community to create artificial intelligence,” Jazayeri says. “The relevance of these models to neurobiology hinges on their ability to additionally capture the inner workings of the brain. The fact that Aran’s model predicts neural data is really important as it suggests that we may be getting closer to building artificial systems that emulate natural intelligence.”

Navigating the world

The study led by Khona, Schaeffer, and Fiete focused on a type of specialized neurons known as grid cells. These cells, located in the entorhinal cortex, help animals to navigate, working together with place cells located in the hippocampus.

While place cells fire whenever an animal is in a specific location, grid cells fire only when the animal is at one of the vertices of a triangular lattice. Groups of grid cells create overlapping lattices of different sizes, which allows them to encode a large number of positions using a relatively small number of cells.

In recent studies, researchers have trained supervised neural networks to mimic grid cell function by predicting an animal’s next location based on its starting point and velocity, a task known as path integration. However, these models hinged on access to privileged information about absolute space at all times — information that the animal does not have.

Inspired by the striking coding properties of the multiperiodic grid-cell code for space, the MIT team trained a contrastive self-supervised model to both perform this same path integration task and represent space efficiently while doing so. For the training data, they used sequences of velocity inputs. The model learned to distinguish positions based on whether they were similar or different — nearby positions generated similar codes, but further positions generated more different codes.

“It’s similar to training models on images, where if two images are both heads of cats, their codes should be similar, but if one is the head of a cat and one is a truck, then you want their codes to repel,” Khona says. “We’re taking that same idea but applying it to spatial trajectories.”

Once the model was trained, the researchers found that the activation patterns of the nodes within the model formed several lattice patterns with different periods, very similar to those formed by grid cells in the brain.

“What excites me about this work is that it makes connections between mathematical work on the striking information-theoretic properties of the grid cell code and the computation of path integration,” Fiete says. “While the mathematical work was analytic — what properties does the grid cell code possess? — the approach of optimizing coding efficiency through self-supervised learning and obtaining grid-like tuning is synthetic: It shows what properties might be necessary and sufficient to explain why the brain has grid cells.”

The research was funded by the K. Lisa Yang ICoN Center, the National Institutes of Health, the Simons Foundation, the McKnight Foundation, the McGovern Institute, and the Helen Hay Whitney Foundation.

Four McGovern Investigators receive NIH BRAIN Initiative grants

by Julie Pryor |

In the human brain, 86 billion neurons form more than 100 trillion connections with other neurons at junctions called synapses. Scientists at the McGovern Institute are working with their collaborators to develop technologies to map these connections across the brain, from mice to humans.

Today, the National Institutes of Health (NIH) announced a new program to support research projects that have the potential to reveal an unprecedented and dynamic picture of the connected networks in the brain. Four of these NIH-funded research projects will take place in McGovern labs.

BRAIN Initiative

In 2013, the Obama administration announced the Brain Research Through Advancing Innovative Neurotechnologies® (BRAIN) Initiative, a public-private research effort to support the development and application of new technologies to understand brain function.

Today, the NIH announced its third project supported by the BRAIN Initiative, called BRAIN Initiative Connectivity Across Scales (BRAIN CONNECTS). The new project complements two previous large-scale projects, which together aim to transform neuroscience research by generating wiring diagrams that can span entire brains across multiple species. These detailed wiring diagrams can help uncover the logic of the brain’s neural code, leading to a better understanding of how this circuitry makes us who we are and how it could be rewired to treat brain diseases.

BRAIN CONNECTS at McGovern

The initial round of BRAIN CONNECTS awards will support researchers at more than 40 university and research institutions across the globe with 11 grants totaling $150 million over five years. Four of these grants have been awarded to McGovern researchers Guoping Feng, Ila Fiete, Satra Ghosh, and Ian Wickersham, whose projects are outlined below:

BRAIN CONNECTS: Comprehensive regional projection map of marmoset with single axon and cell type resolution
Team: Guoping Feng (McGovern Institute, MIT), Partha Mitra (Cold Spring Harbor Laboratory), Xiao Wang (Broad Institute), Ian Wickersham (McGovern Institute, MIT)

Summary: This project will establish an integrated experimental-computational platform to create the first comprehensive brain-wide mesoscale connectivity map in a non-human primate (NHP), the common marmoset (Callithrix jacchus). It will do so by tracing axonal projections of RNA barcode-identified neurons brain-wide in the marmoset, utilizing a sequencing-based imaging method that also permits simultaneous transcriptomic cell typing of the identified neurons. This work will help bridge the gap between brain-wide mesoscale connectivity data available for the mouse from a decade of mapping efforts using modern techniques and the absence of comparable data in humans and NHPs.

BRAIN CONNECTS: A center for high-throughput integrative mouse connectomics
Team: Jeff Lichtman (Harvard University), Ila Fiete (McGovern Institute, MIT), Sebastian Seung (Princeton University), David Tank (Princeton University), Hongkui Zeng (Allen Institute), Viren Jain (Google), Greg Jeffries (Oxford University)

Summary: This project aims to produce a large-scale synapse-level brain map (connectome) that includes all the main areas of the mouse hippocampus. This region is of clinical interest because it is an essential part of the circuit underlying spatial navigation and memory and the earliest impairments and degeneration related to Alzheimer’s disease.

BRAIN CONNECTS: The center for Large-scale Imaging of Neural Circuits (LINC)
Team: Anastasia Yendiki (MGH), Satra Ghosh (McGovern, MIT), Suzanne Haber (University of Rochester), Elizabeth Hillman (Columbia University)

Summary: This project will generate connectional diagrams of the monkey and human brain at unprecedented resolutions. These diagrams will be linked both to the neuroanatomic literature and to in vivo neuroimaging techniques, bridging between the rigor of the former and the clinical relevance of the latter. The data to be generated by this project will advance our understanding of brain circuits that are implicated in motor and psychiatric disorders, and that are targeted by deep-brain stimulation to treat these disorders.

BRAIN CONNECTS: Mapping brain-wide connectivity of neuronal types using barcoded connectomics
Team: Xiaoyin Chen (Allen Institute), Ian Wickersham (McGovern Institute, MIT), and Justus Kebschull of JHU

Summary: This project aims to optimize and develop barcode sequencing-based neuroanatomical techniques to achieve brain-wide, high-throughput, highly multiplexed mapping of axonal projections and synaptic connectivity of neuronal types at cellular resolution in primate brains. The team will work together to apply these techniques to generate an unprecedented multi-resolution map of brain-wide projections and synaptic inputs of neurons in the macaque visual cortex at cellular resolution.

 

Ila Fiete wins Swartz Prize for Theoretical and Computational Neuroscience

by MIT Dept of Brain and Cognitive Sciences |

The Society for Neuroscience (SfN) has awarded the Swartz Prize for Theoretical and Computational Neuroscience to Ila Fiete, professor in the Department of Brain and Cognitive Sciences, associate member of the McGovern Institute for Brain Research, and director of the K. Lisa Yang Integrative Computational Neuroscience Center. The SfN, the world’s largest neuroscience organization, announced that Fiete received the prize for her breakthrough research modeling hippocampal grid cells, a component of the navigational system of the mammalian brain.

“Fiete’s body of work has already significantly shaped the field of neuroscience and will continue to do so for the foreseeable future,” states the announcement from SfN.

“Fiete is considered one of the strongest theorists of her generation who has conducted highly influential work demonstrating that grid cell networks have attractor-like dynamics,” says Hollis Cline, a professor at the Scripps Research Institute of California and head of the Swartz Prize selection committee.

Grid cells are found in the cortex of all mammals. Their unique firing properties, creating a neural representation of our surroundings, allow us to navigate the world. Fiete and collaborators developed computational models showing how interactions between neurons can lead to the formation of periodic lattice-like firing patterns of grid cells and stabilize these patterns to create spatial memory. They showed that as we move around in space, these neural patterns can integrate velocity signals to provide a constantly updated estimate of our position, as well as detect and correct errors in the estimated position.

Fiete also proposed that multiple copies of these patterns at different spatial scales enabled efficient and high-capacity representation. Next, Fiete and colleagues worked with multiple collaborators to design experimental tests and establish rare evidence that these pattern-forming mechanisms underlie the function of memory pattern dynamics in the brain.

“I’m truly honored to receive the Swartz Prize,” says Fiete. “This prize recognizes my group’s efforts to decipher the circuit-level mechanisms of cognitive functions involving navigation, integration, and memory. It also recognizes, in its focus, the bearing-of-fruit of dynamical circuit models from my group and others that explain how individually simple elements combine to generate the longer-lasting memory states and complex computations of the brain. I am proud to be able to represent, in some measure, the work of my incredible students, postdocs, collaborators, and intellectual mentors. I am indebted to them and grateful for the chance to work together.”

According to the SfN announcement, Fiete has contributed to the field in many other ways, including modeling “how entorhinal cortex could interact with the hippocampus to efficiently and robustly store large numbers of memories and developed a remarkable method to discern the structure of intrinsic dynamics in neuronal circuits.” This modeling led to the discovery of an internal compass that tracks the direction of one’s head, even in the absence of external sensory input.

“Recently, Fiete’s group has explored the emergence of modular organization, a line of work that elucidates how grid cell modularity and general cortical modules might self-organize from smooth genetic gradients,” states the SfN announcement. Fiete and her research group have shown that even if the biophysical properties underlying grid cells of different scale are mostly similar, continuous variations in these properties can result in discrete groupings of grid cells, each with a different function.

Fiete was recognized with the Swartz Prize, which includes a $30,000 award, during the SfN annual meeting in San Diego.

Other recent MIT winners of the Swartz Prize include Professor Emery Brown (2020) and Professor Tomaso Poggio (2014).

Study urges caution when comparing neural networks to the brain

Anne Trafton |

Neural networks, a type of computing system loosely modeled on the organization of the human brain, form the basis of many artificial intelligence systems for applications such speech recognition, computer vision, and medical image analysis.

In the field of neuroscience, researchers often use neural networks to try to model the same kind of tasks that the brain performs, in hopes that the models could suggest new hypotheses regarding how the brain itself performs those tasks. However, a group of researchers at MIT is urging that more caution should be taken when interpreting these models.

In an analysis of more than 11,000 neural networks that were trained to simulate the function of grid cells — key components of the brain’s navigation system — the researchers found that neural networks only produced grid-cell-like activity when they were given very specific constraints that are not found in biological systems.

“What this suggests is that in order to obtain a result with grid cells, the researchers training the models needed to bake in those results with specific, biologically implausible implementation choices,” says Rylan Schaeffer, a former senior research associate at MIT.

Without those constraints, the MIT team found that very few neural networks generated grid-cell-like activity, suggesting that these models do not necessarily generate useful predictions of how the brain works.

Schaeffer, who is now a graduate student in computer science at Stanford University, is the lead author of the new study, which will be presented at the 2022 Conference on Neural Information Processing Systems this month. Ila Fiete, a professor of brain and cognitive sciences and a member of MIT’s McGovern Institute for Brain Research, is the senior author of the paper. Mikail Khona, an MIT graduate student in physics, is also an author.

Ila Fiete leads a discussion in her lab at the McGovern Institute. Photo: Steph Stevens

Modeling grid cells

Neural networks, which researchers have been using for decades to perform a variety of computational tasks, consist of thousands or millions of processing units connected to each other. Each node has connections of varying strengths to other nodes in the network. As the network analyzes huge amounts of data, the strengths of those connections change as the network learns to perform the desired task.

In this study, the researchers focused on neural networks that have been developed to mimic the function of the brain’s grid cells, which are found in the entorhinal cortex of the mammalian brain. Together with place cells, found in the hippocampus, grid cells form a brain circuit that helps animals know where they are and how to navigate to a different location.

Place cells have been shown to fire whenever an animal is in a specific location, and each place cell may respond to more than one location. Grid cells, on the other hand, work very differently. As an animal moves through a space such as a room, grid cells fire only when the animal is at one of the vertices of a triangular lattice. Different groups of grid cells create lattices of slightly different dimensions, which overlap each other. This allows grid cells to encode a large number of unique positions using a relatively small number of cells.

This type of location encoding also makes it possible to predict an animal’s next location based on a given starting point and a velocity. In several recent studies, researchers have trained neural networks to perform this same task, which is known as path integration.

To train neural networks to perform this task, researchers feed into it a starting point and a velocity that varies over time. The model essentially mimics the activity of an animal roaming through a space, and calculates updated positions as it moves. As the model performs the task, the activity patterns of different units within the network can be measured. Each unit’s activity can be represented as a firing pattern, similar to the firing patterns of neurons in the brain.

In several previous studies, researchers have reported that their models produced units with activity patterns that closely mimic the firing patterns of grid cells. These studies concluded that grid-cell-like representations would naturally emerge in any neural network trained to perform the path integration task.

However, the MIT researchers found very different results. In an analysis of more than 11,000 neural networks that they trained on path integration, they found that while nearly 90 percent of them learned the task successfully, only about 10 percent of those networks generated activity patterns that could be classified as grid-cell-like. That includes networks in which even only a single unit achieved a high grid score.

The earlier studies were more likely to generate grid-cell-like activity only because of the constraints that researchers build into those models, according to the MIT team.

“Earlier studies have presented this story that if you train networks to path integrate, you’re going to get grid cells. What we found is that instead, you have to make this long sequence of choices of parameters, which we know are inconsistent with the biology, and then in a small sliver of those parameters, you will get the desired result,” Schaeffer says.

More biological models

One of the constraints found in earlier studies is that the researchers required the model to convert velocity into a unique position, reported by one network unit that corresponds to a place cell. For this to happen, the researchers also required that each place cell correspond to only one location, which is not how biological place cells work: Studies have shown that place cells in the hippocampus can respond to up to 20 different locations, not just one.

When the MIT team adjusted the models so that place cells were more like biological place cells, the models were still able to perform the path integration task, but they no longer produced grid-cell-like activity. Grid-cell-like activity also disappeared when the researchers instructed the models to generate different types of location output, such as location on a grid with X and Y axes, or location as a distance and angle relative to a home point.

“If the only thing that you ask this network to do is path integrate, and you impose a set of very specific, not physiological requirements on the readout unit, then it’s possible to obtain grid cells,” says Fiete, who is also the director of the K. Lisa Yang Integrative Computational Neuroscience Center at MIT. “But if you relax any of these aspects of this readout unit, that strongly degrades the ability of the network to produce grid cells. In fact, usually they don’t, even though they still solve the path integration task.”

Therefore, if the researchers hadn’t already known of the existence of grid cells, and guided the model to produce them, it would be very unlikely for them to appear as a natural consequence of the model training.

The researchers say that their findings suggest that more caution is warranted when interpreting neural network models of the brain.

“When you use deep learning models, they can be a powerful tool, but one has to be very circumspect in interpreting them and in determining whether they are truly making de novo predictions, or even shedding light on what it is that the brain is optimizing,” Fiete says.

Kenneth Harris, a professor of quantitative neuroscience at University College London, says he hopes the new study will encourage neuroscientists to be more careful when stating what can be shown by analogies between neural networks and the brain.

“Neural networks can be a useful source of predictions. If you want to learn how the brain solves a computation, you can train a network to perform it, then test the hypothesis that the brain works the same way. Whether the hypothesis is confirmed or not, you will learn something,” says Harris, who was not involved in the study. “This paper shows that ‘postdiction’ is less powerful: Neural networks have many parameters, so getting them to replicate an existing result is not as surprising.”

When using these models to make predictions about how the brain works, it’s important to take into account realistic, known biological constraints when building the models, the MIT researchers say. They are now working on models of grid cells that they hope will generate more accurate predictions of how grid cells in the brain work.

“Deep learning models will give us insight about the brain, but only after you inject a lot of biological knowledge into the model,” Khona says. “If you use the correct constraints, then the models can give you a brain-like solution.”

The research was funded by the Office of Naval Research, the National Science Foundation, the Simons Foundation through the Simons Collaboration on the Global Brain, and the Howard Hughes Medical Institute through the Faculty Scholars Program. Mikail Khona was supported by the MathWorks Science Fellowship.