Unsupervised Thinking

You may have heard of predictive coding; it's a theory that gets around. In fact, it's been used to understand everything from the retina to consciousness. So, before we get into the details, we start this episode by describing our impressions of predictive coding. Where have we encountered it? Has it influenced our work? Why do philosophers like it? And, finally, what does it actually mean? Eventually we settle on a two-tiered definition: "hard" predictive coding refers to a very specific hypothesis about how the brain calculates errors, and "soft" predictive coding refers to the general idea that the brain predicts things. We then get into how predictive coding relates to other theories, like Bayesian modeling. But like Bayesian models, which we've covered on a previous episode, predictive coding is prone to "just-so" stories. So we discuss what concrete predictions predictive coding can make, and whether the data supports them. Finally, Grace tries to describe the free energy principle, which extends predictive coding into a grand unified theory of the brain and beyond.

There is a tension when it comes to the study of behavior in neuroscience. On the one hand, we would love to understand animals as they behave in the wild---with the full complexity of the stimuli they take in and the actions they emit. On the other hand, such complexity is almost antithetical to the scientific endeavor, where control over inputs and precise measurement of outputs is required. Throw in the constraints that come when trying to record from and manipulate neurons and you've got a real mess. In this episode, we discuss these tensions and the modern attempts to resolve them.

First, we take the example of decision-making in rodents to showcase what behavior looks like in neuroscience experiments (and how strangely we use the term "decision-making"). In these studies, using more natural stimuli can help with training and lead to better neural responses. But does going natural make the analysis of the data more difficult? We then talk about how machine learning can be used to automate the analysis of behavior, and potentially remove harmful human biases. Throughout, we provide multiple definitions of "behavior", Grace relates animal training to parenting, and our special guest Adam Calhoun uses his encyclopedic knowledge of this area to provide many insightful examples!

On this unique episode of Unsupervised Thinking, we talk not about a particular area of science, but about the process of doing science itself. In particular, we're discussing how scientists take in information from their niche research areas and beyond. The topic for this free-form conversation stemmed from interviews we collected at the latest Computational and Systems Neuroscience Conference (Cosyne), where we asked people to tell us about a research finding from outside their area that they thought was cool. You'll hear those interviews in this episode, along with our motivation for asking that question and our reaction to the responses. We then go on to speak broadly about our experiences at different conferences both big and small. In particular, we reveal how attending talks from far-reaching areas of science is a great way to build appreciation for your field and contextual your research. Ultimately, influence from talks and colleagues is how scientists choose their projects, and so decisions of what to consume can have long-lasting effects. We give our personal examples of times when talks have unexpectedly impacted our research, and the concrete things we do to keep up with the literature. Throughout, there is also talk of the culture divides that arise in research and Josh tells us about Game of Thrones special effects are done. Thanks to all the Cosyne attendees who agreed to be interviewed for this, and special thanks to Nancy Padilla, our special guest for this episode!

Sleep is such a ubiquitous part of our lives we may forget just how weird of a thing it is to spend a third of our days laying in darkness. In this episode on the science of sleep, we start by describing types of sleep (while appreciating its strangeness) and the negative cognitive effects of missing out on it. We also discuss the potential role of rapid eye movement (REM) sleep in training neural connections and how that idea has been ported to artificial intelligence. We then take a rare (for us) trip to the brain stem to understand the neural mechanisms of switching from wake to sleep and nonREM to REM sleep. While there we appreciate how difficult it is to tease apart the function of different neuron types in a heavily redundant system. Finally, we dip our toes into the world of dreams, discussing a theory of dreaming's role in developing consciousness and our own experiences with lucid dreaming. Throughout, we go down tangents to argue about the value of speculative science and how to quantify knowledge via Google search results.

In the late 19th century, many of the most basic foundations of neuroscience were laid, but not without a fight. In this episode, we cover the famous debate between Camillo Golgi and Santiago Ramon y Cajal regarding the very nature of neurons: are they separate cells or one conjoined mass? We also tell the story of some of the lesser known players, remark on the inability of those on the losing side to let go, and ask what a modern Neuron Doctrine debate would look like. After covering the history, we discuss two modern rethinkings of the Neuron Doctrine that focus the question not just on anatomy, but on whether neurons are the functional unit of the brain. What does "functional unit" mean? We don't know, but as we grapple with it we speak about the need to balance simplification with appreciation of details and (unfortunately) attempt to understand emergence again.

In this very special episode of Unsupervised Thinking, we bring together a group of neuroscientists and neuroscience enthusiasts to have a semi-structured discussion on the past, present, and future of the field of neuroscience. The group includes your three regular hosts plus Yann, Alex, and Ryan (whose voice you may recall from our Deep Learning episode) and we each give our thoughts on what got us into neuroscience, what we feel the field is lacking, and where the field will be in 20 years. This leads us on a path of discussing statistics, emergence, religion, depression, behavior, engineering, society, and more!

In the second part of our discussion on the concept of coding, we get into specific examples of what could count as the "neural code". In particular, we talk rate vs temporal coding along with examples of each in different organisms. We then get into the use of coding as a metaphor. How does it influence thought and experiments? Is an "action" or "process" metaphor a better alternative? We also discuss why language is the wrong metaphor for the neural code and the relationship between metaphors and formal mathematical systems. To illustrate the tension between a coding framework versus an action/process framework, we discuss a concrete example where the same finding is explained two different ways: does orientation-selective cells in primary visual cortex arise from sparse coding principles or is it just what happens when you train a system to do vision?

We read:

Neural Representation and the Cortical Code

Is coding a relevant metaphor for the brain?

The concept of "coding," along with language referencing it, is abundant in neuroscience whether describing sensory systems, cognition, or motor control. The notion that neurons encode information is so core to neuroscience it is almost difficult to imagine the field without it. In the first part of this two-part discussion on coding, we talk about the origins of the coding concept. We start with some of the early experimental work that demonstrated the most basic response properties of neurons. We then delve into a conference report from the 60's that summarized the state of the field at that time and find it (depressingly?) relevant for today. In particular, the focus at that time on coding as only an imperfect metaphor is contrasted with its perhaps outsized role in modern work. We ask things like: When is the metaphor working, and when is it stretched beyond recognition? What qualities does a code need to have to be a candidate for the "neural code"? and If there is a neural code, who's reading it out? By the end, Grace freaks out about how abstract "information processing" is, and Josh and Conor claim the whole world is just one big information processing machine.

We read:

Neurosciences Research Program report on Neural Coding

And mentioned:

List of recommended readings on coding

Spikes

Our fMRI episode

Upcoming readings for Part 2:

Neural Representation and the Cortical Code

Is coding a relevant metaphor for the brain?

You move your eyes several times a second, making choices about what to attend to without even noticing. That is a lot of behavioral data that scientists could use to understand underlying computations, preferences, memories, and intentions. On this episode, we talk about just such endeavors to understand cognition by monitoring eye movements. Eye movements are a comparatively easy thing to measure (though as we

discuss, some of the older methods seemed like torture devices), and can be used in a wide range of settings: in animals and babies, in the lab or in the wild. In this episode, Josh regals us with tales of using a modern eye-tracking device at a conference, and we talk about the basic findings you can discover from that and exactly how surprising or interesting they are. We then get into how tasks influence eye movements and the many forms of memory that eye movements can measure, including differences between novices and experts. We top it all off with a delightful study about looking at porn, and some speculations about how eye-tracking could be used in the future.

We read:

Eye movements in natural behavior

Worth a Glance: Using Eye Movements to Investigate the Cognitive Neuroscience of Memory

Sex differences in viewing sexual stimuli: An eye-tracking study in men and women

A Breadth-First Survey of Eye Tracking Applications

For this episode, special guest Nancy Padilla (E5: Neural Oscillations) returns to talk about a topic she now studies: social neuroscience. We get into the methods this rather new field uses to probe the neural processes behind social interaction, including the inherent difficulties in studying such a complex subject. We go from special pathways for bottom-up social processing such as smells and facial recognition to ideas about theory of mind and cooperation. In the process, we hit on the mirror neuron system, simultaneous dual-brain recordings, and the role of philosophy in the field. Two common questions throughout are: (1) are humans unique amongst animals? and (2) is social processing unique amongst neuroscience topics? Ultimately, we try to discover if social neuroscience is greater than the sum of its parts.

We read:

Conceptual Challenges and Directions for Social Neuroscience

Brain Basis of Human Social Interaction: From Concepts to Brain Imaging

Brain-to-Brain coupling: A mechanism for creating and sharing a social world

And mentioned:

Episode 10: Brain Size