(1) How can we find out how the brain works?

It’s hard to discover specific new facts about brains — how synaptic strengths change, how a population of neurons’ activity changes over seconds, how brain networks change over weeks and decades. But the biggest challenge for understanding “how the brain works” is making the right bridges across levels of analysis. A satisfying explanation will have to show how phenomena at each higher level can be explained in terms of the operation of components at the next lower level. These include how thoughts and behaviors can be understood in terms of activity in neural populations; and how such aggregate activity is composed of the biophysical structure of neurons. I expect that progress will likely be made once both “sides” of the bridge can be expressed mathematically, that is, when we have a formal quantitative vocabulary that can capture the relevant properties of thoughts and of neural populations; or of neural populations and of neuron biophysics. In my view, a critical open question is: will principles developed for one part of cognition (e.g. object recognition) generalize rather well, or rather badly, to other parts of cognition (e.g. language understanding)?

(2) What will your talk at CCN 2017 be about?

How we understand others’ emotions. Humans observers can recognize and reason about highly-differentiated, or fine-grained, emotions. We recognize anxiety in our friends, distinguish their anxiety from their disappointment or regret, and try to respond in appropriate ways; but how do we make such specific and accurate emotion attributions to another person? I will argue that human emotion understanding does not rely (much) on sensitive perception of dynamic emotional expressions in faces, bodies or voices. Rather, human emotion understanding depends on understanding the “context” of a person’s emotional expressions. And I will argue that what “context” means here, is a representation of how events and outcomes affect her desires, constraints, values, and causal power. Thus, a computational model of emotion understanding requires a model of intuitive reasoning about others’ mental states. I will sketch our plans for such a model. And in so doing, I will give an example of the potential for computational cognitive neuroscience: combining computational models of cognition, with computational techniques for analysis of neuroscientific data.

(3) How can cognitive science, computational neuroscience, and artificial intelligence best work together?

The most important challenge for these inter-disciplinary efforts is to develop strong bridging hypotheses and proposals for how to connect phenomena across levels of analyses. In the end, we will only understand the human mind when we can describe its phenomena, in quantitative detail, at each level of analysis (society, behavior, systems, cells), but also can articulate how the levels relate to each other. Bridging hypotheses are like a dictionary, translating between different languages.

I hope that in the future, cognitive science will describe a horizon of the kinds of problems worth solving. Cognitive scientists work on the kinds of problems that need to be solved not only to understand how human minds work, but also to create machines that function in a human world.

(4) What current developments are you most excited about?

I am excited about the push to study increasingly complex cognitive (and social) phenomena in rodent models, bringing the scale of that research to hard problems in cognition. I am excited about the development of tools to probe the ‘code’ of neural populations through dense recording. I am excited about the push for replicability and transparency in fMRI research. And I am excited about the emerging cognitive neuroscience of human infant brains — which I am working on! But probably won’t talk about this year.

(5) What do you hope to learn at CCN 2017?

I am very excited for CCN 2017 — more than I’ve been for a conference in quite a while! I wish I could be there to hear all of the talks, because the list of speakers looks so exciting.

(1) How can we find out how the brain works?

I am primarily interested in understanding human intelligence, and from that point of view, I am worried about the massive move towards rodent models in neuroscience. While the abundance of available techniques makes this move attractive, there is not much evidence that rodents plan strategically, reflect on their own decisions, infer the mental states of others, weigh the value of information against costs, etc. One way forward would be to demonstrate these abilities in rodents. An alternative would be to become less obsessed by techniques.

At a more sociological level, I am old-fashioned and strongly believe in small, hypothesis-driven science. While some problems in neuroscience might be best addressed using big data, big simulations, or big collaborations, my sense is that those currently involve more hype than substance. Neuroscience and cognitive science have come far with a “letting a hundred flowers bloom” approach, and there is no evidence that this approach is bankrupt. More specifically in computational neuroscience, small science often amounts to a search for evolutionarily meaningful organizing principles, perhaps initially in a toy model – this is my favorite approach.

(2) What will your talk at CCN 2017 be about?

It will be a tutorial on mathematical modeling of behavior. I will try to categorize such models, dig a bit deeper into Bayesian models, and spend a good amount of time on best practices for model fitting and model comparison. I am very excited that CCN offers tutorials and takes them seriously; perhaps in future editions, we can add more time for doing exercises?

(3) How can cognitive science, computational neuroscience, and artificial intelligence best work together?

One missing piece in the field is the psychological plausibility of neural network models of cortex. There is a lot of attention for biological plausibility, and for linking to neural data. However, attempts to link to behavioral data typically stay extremely crude, of the type “humans can do the task and the network can do the task”, or barely better, “the network approaches human accuracy on this task”. We should use more tools from basic psychophysics to interrogate and challenge neural networks: subject a trained network to a psychophysical experiment in which stimuli are parametrically varied, and compare its psychometric curves quantitatively to human data. Bonus: compare the same behavioral-level models on the network output and compare model goodness-of-fit ranking to that obtained from human data. Emphasizing psychological plausibility of networks could help to unify the fields represented at CCN.

(4) What current developments are you most excited about?

Connections between behavioral economics, reinforcement learning, and inference: e.g. information gathering as an economic problem, keeping track of uncertainty in RL, and thinking ahead in sequential decision tasks with large decision trees.

(5) What do you hope to learn at CCN 2017?

I am hoping to find a new conference home, after years of mild frustration that Cosyne has too little interest in behavior and cognition, and that VSS has too little computational modeling and machine learning.

(1) How can we find out how the brain works?

It all begins with thoughtful descriptions of the computations that the brain solves, which are often directly reflected in behavior. Ultimately, a description of “how” is formalized by a model that provides a non-trivial account of data. Crucially, while many of us have been taught that the ultimate test of understanding something is to build it, recent work in our field highlights that you can build something without deeply understanding how it works. Model interpretability is one of the biggest challenges that we currently face.

(2) What will your talk at CCN 2017 be about?

Single-trial visual memory. I am fascinated by our remarkable ability to remember the objects and scenes that we have encountered previously, after viewing thousands of images, and by the fact that we remember these images with considerable detail. How do our brains manage this? This question is an excellent case study for CCN, because its answer is interesting at so many different levels. Algorithmically – what are the learning rules responsible for visual memory storage and how does the brain manage perceptual stability in the face of single-trial plasticity? Theoretically – what is a memory tag that indicates “I’ve seen this before” useful for? And is it a natural byproduct of a system optimized to predict? And biophysically – while I’ve never really been drawn toward detailed biophysical descriptions of information processing (e.g. visual processing), I find them much more compelling when contemplating memory. In the case of information processing, spikes are the primary currency, and it is thus useful to discuss how abstract quantities such as ‘signals’ are transformed as information propagates from one stage to the next. In the case of information storage, spikes can signal memory recollection, but spikes are not themselves the memory.

(3) How can cognitive science, computational neuroscience, and artificial intelligence best work together?

One of the most important factors is clear communication – we all work with sophisticated and complex ideas, and if I can’t understand what you are talking about, I can’t adopt what you have figured out. We all need to invest time and energy into the explanation as well as the discovery. Interdisciplinary meetings – like CCN! – facilitate this. As does cross-field collaboration, whether it’s writing a review article or co-mentoring a trainee.

(4) What current developments are you most excited about?

New developments in biotechnology that allow us to monitor and manipulate neural activity in incredibly specific ways (e.g. optogenetics). These tools are revolutionizing our field, because they allow for causal tests of our descriptions of neural computation like never before. For example, these tools have been used in mice to tag the specific subpopulations of neurons activated during memory formation such that those neurons can be perturbed in purposeful ways later on.

(5) What do you hope to learn at CCN 2017?

At any meeting, I am usually most excited to discover great work that I was previously unaware of. Given the convergence of different fields at CCN, I am looking forward to learning about new ideas and approaches that I currently know little about.

Photo by Jeroen Oerlemans

(1) How can we find out how the brain works?

I believe that formal models that make simultaneous predictions about different modalities such as behavior and the brain are powerful tools. Such tools could help to gain a better mechanistic understanding of brain function.

(2) What will your talk at CCN 2017 be about?

In my talk I will focus on the human subcortex. I will show how different scales can be combined including information from the cellular level by means of immunohistochemistry to neurocognitive modeling and how they can be combined in joint models. Finally, I will show how this knowledge can directly translate to the bedside.

(3) How can cognitive science, computational neuroscience, and artificial intelligence best work together?

Organizing exciting new meetings such as CCN is an excellent start. In addition, hands-on workshops are an interesting format to get people familiarized with the tools and data each discipline has to offer. In the lab, it is fun to have people with backgrounds from the cognitive sciences, basic/cognitive neurosciences, and artificial intelligence work together and see how their curiosity drives them to learn with and from each other.

(4) What current developments are you most excited about?

I am excited about technologies that give us noninvasive access to small structures that lie deep in the brain including ultra-high field MRI and Connectom scanners. My hope is that this technology will deliver better brain data, which in turn is essential to developing more precise neurocognitive models and may ultimately translate to the bedside.

(5) What do you hope to learn at CCN 2017?

A lot.

(1) How can we find out how the brain works?

By testing neurobiological hypotheses that address the “how” question at a variety of levels. To me, “how the brain works” is a biological problem because I am less interested in mimicking the brain with a machine than I am in assessing what goes wrong when the brain doesn’t work, and how we might remedy the fault. To this end, functional equivalence (like airplanes to birds), which might interest the engineer, is not enough and possibly detrimental—a misguided diversion. To make progress on the “how” of cognitive function, my approach is to focus less on the representation of information and more on what the organism does with the information. To put it crudely, start at the motor system and work backwards. Another practical guide is the time signature of neural processing. Propositions, beliefs, plans and decisions transpire on time scales that necessitate persistence—that are free from the immediacy of sensory processing and online motor control. We need big (new) ideas, not big data acquired under the presumption that all the brain has to work with is connectionism and Hebbian plasticity.

(2) What will your talk at CCN 2017 be about?

I haven’t decided yet. Yes, I will actually talk about that.

(3) How can cognitive science, computational neuroscience, and artificial intelligence best work together?

By drawing on and attempting to influence the missing element from this list: experimental neurobiology. Abandon functionalism (think airplane/bird neural network/brain), magical incantations (add fairy dust du jour and the gorilla is now consciously perceived, where dust equals oscillation or synchrony or ignition), parlor tricks (e.g., chips that [appear to] control limbs or algorithms that read minds), and obfuscation (e.g., appeals to high dimensional representations without a testable/plausible notion of a biological mechanism for read out—unless we count the statistician with a computer as a biological mechanism).

(4) What current developments are you most excited about?

The emerging intersection of circuit- and systems-level approaches in behavior. The ability to record from neurons based on their connectivity to other neurons in nonhuman primates—still just a promise, but it will open a new era in neurophysiology.

(5) What do you hope to learn at CCN 2017?

Why 90% of what I just wrote is wrong. And/or that there are a few sisters and brothers out there who share similar prejudices. And/or whether my kevlar vest works as advertised (just in case, I’m A positive).

1 How can we find out how the brain works?

We will need to continue to study the brain at multiple scales, both at the neuronal level, at the macro level (via brain imaging), and at the behavioral level. And we need to continue to bring these worlds together. We also need to start pushing brain imaging experiments in humans out into the real world. We can learn something about human language understanding by watching people read single words or single sentences, but we will miss out on the higher level comprehension areas that are required for larger scale understanding and reasoning. Similarly, viewing pictures or watching videos tells us something about vision, but interacting with objects in the real world will likely tell us more. There is tremendous value in tightly controlled experimental paradigms, but we also need some people doing the hard work that gets at the more holistic aspects of brain information processing.

2 What will your talk at CCN 2017 be about?

My tutorial talk is broadly about AI, but I will give several examples of how we have programmed computers to abstractly represent the world, and how that abstract representation relates to people’s understanding of the world (measured with brain imaging).

3 How can cognitive science, computational neuroscience, and artificial intelligence best work together?

By training people to straddle these areas. I was incredibly lucky to get my training both in machine learning and neuroscience, and it has allowed me to live in (and contribute to) both areas. I hope my students will go on to do the same thing.

4 What current developments are you most excited about?

I’m very excited to see how the hidden representations learned by artificial neural networks are being related to brain images for people viewing the same stimuli. This is incredibly exciting, and I think hints that the two areas are soon to collide.

5 What do you hope to learn at CCN 2017?

You’ve put together an amazing program and I’m excited to hear from Rebecca Saxe and Yael Niv.  The panel discussions have great groups of people and could be very interesting!

1 How can we find out how the brain works?

As a cognitive scientist I normally think about this question from the perspective of Marr’s levels of analysis. Understanding how the brain works is a question at what Marr called the “implementation” level, but I think a lot of insight can be gained by asking why the brain does what it does — a question at Marr’s “computational” level. Between those levels of analysis is the “algorithmic” level, which looks at the particular cognitive processes that are involved in solving a problem. Over the last few years a lot of progress has been made at both the implementation level and the computational level, but I think the algorithmic level gets neglected when we think about the brain. Understanding the algorithms that brains execute — and how brains learn to execute those algorithms — is going to be a critical part of finding out how the brain works.

2 What will your talk at CCN 2017 be about?

I will talk about a formal framework for linking these different levels of analysis. Computational-level explanations focus on the structure of abstract problems and their ideal solutions, explaining behavior in terms of optimal solutions to the problems that we face in everyday life. However, finding those optimal solutions can be extremely computationally expensive. So, my collaborators and I have started to look at the consequences of introducing resource constraints: if you know you have to solve a problem using a particular computational architecture, with associated computational costs, how should you solve the problem? This “bounded optimality” approach pushes the principle of optimization down to the algorithmic level, and gives us a new set of tools for exploring some classic questions about human cognition.

3 How can cognitive science, computational neuroscience, and artificial intelligence best work together?

I think these three approaches need to work together to give explanations of how the mind works all levels of analysis. Artificial intelligence can help us find good solutions to the kinds of problems that intelligence organisms face, providing the elements we need for constructing computational-level theories. Cognitive science puts these theories together with behavior to really understand the cognitive processes that people engage in. Computational neuroscience helps us understand how those cognitive processes could be executed by neurons. However, explanations at each of these levels can inform one another. For example, there is a productive feedback loop between AI and cognitive science, with AI leading to new theories of human cognition and cognitive science highlighting the ways in which AI still falls short. I think similar feedback loops can be constructed between all three disciplines.

4 What current developments are you most excited about?

The three things I am currently most excited about are understanding how people are so good at finding strategies for solving complex problems (something that is still beyond the limits of current AI), the potential for crowdsourcing technology to be transformative for the way that we conduct behavioral research (I think it is currently massively underutilized), and exploiting the advances in deep learning for representing complex stimuli such as images and text in order to study human cognition with naturalistic stimuli.

5 What do you hope to learn at CCN 2017?

I’m really interested in those feedback loops between different disciplines, so I hope to learn more about how machine learning and AI have been influencing contemporary neuroscience.

1 How can we find out how the brain works?

If there is a compact description of the computational principles which explain how the brain manages to provide us with our intelligence, this is something I would consider the core explanation for how the brain works – a little bit like the laws of physics for our physical world. Note that this is very different from the structured observation of our world in all its encyclopedic detail, which provides a useful map of our world, but not a principled explanation. Just replace ‘world’ by ‘brain’. My thesis is that those principles would also allow us to build intelligent machines and that at the heart of our intelligence is our ability to learn and make sense of the world, by observing it and interacting with it. That is why I believe in the importance of a continuous discussion between the brain researchers and AI researchers, especially those in machine learning – particularly deep learning and neural networks. This is likely to benefit AI research as well, as it has in the past.

2 What will your talk at CCN 2017 be about?

I will start by discussing the recent progress in deep learning research, focusing on the inspiration from neuroscience and cognition. I will also talk about novel work, which aims at bridging the gap between back-propagation, the workhorse of deep learning, and neuroscience, as well as about unsupervised reinforcement learning approaches on the cognitive side, which may help to formalize in a machine learning framework how a learner could discover the notions of attributes and objects (as independently controllable aspects of the environment).

3 How can cognitive science, computational neuroscience, and artificial intelligence best work together?

By learning about each other’s developments, collaborating, as usual!

4 What current developments are you most excited about?

It is becoming clear to me that it is very plausible that the brain developed a learning strategy analogous to back-propagation in order to estimate gradients of a training objective, thus addressing the very central question of credit assignment, which is also at the heart of the success of deep learning, in which many areas (layers in a deep neural network) are jointly trained in a coordinated way towards a common objective. Many questions remain open in the quest to bridge this gap between backprop and neuroscience, and we are approaching the time when the questions will be sufficiently precise to be investigated by experimentalists.
I am also very excited about recent developments in machine learning that connect back to older questions raised by classical AI and cognitive science regarding higher-level cognitive notions such as objects, agents, reasoning, memory and knowledge representation, bringing together the expertise in deep learning and in reinforcement learning. There is here again an opportunity for fruitful multi-disciplinary investigations, which could lead us towards a better understanding of cognition – going beyond the current success of deep learning for perception tasks.

5 What do you hope to learn at CCN 2017?

I am hoping to learn about new developments and views in cognitive computational neuroscience, since I do not follow that literature, and as I wrote above, I believe that the potential for positive interactions is quite high.