E-Book, Englisch, 424 Seiten
Reihe: PublishDrive
Louis-Charles Computational Neuroscience Explained
1. Auflage 2026
ISBN: 979-8-90498-706-0
Verlag: PublishDrive
Format: EPUB
Kopierschutz: 0 - No protection
From Neurons, Synapses, and Spiking Circuits to the AI Architectures They Inspired - A Technical Guide for ML Engineers and AI Practitioners
E-Book, Englisch, 424 Seiten
Reihe: PublishDrive
ISBN: 979-8-90498-706-0
Verlag: PublishDrive
Format: EPUB
Kopierschutz: 0 - No protection
Computational Neuroscience Explained is a plain-English engineering and science guide for AI and ML engineers, neuroscientists in training, students, and technically curious readers who want to understand how brains actually compute and how that computation relates to modern artificial intelligence. Set inside the fictional Cortex Lab, the book uses a sustained narrative to carry readers from the biophysics of a single neuron all the way to whole-brain dynamics and the open questions at the frontier of the field.
The opening chapters establish the foundational machinery: the membrane potential, ion channels and the Hodgkin-Huxley equations, the squid giant axon experiments, the McCulloch-Pitts neuron, and how a single biological neuron actually integrates and signals. From there the book turns to synapses, plasticity, and learning rules-LTP and LTD, spike-timing-dependent plasticity, three-factor rules, and the biological story behind the abstract learning rules that drive deep learning.
The middle chapters cover neural coding (rate codes, temporal codes, and population codes), small circuits and network motifs, recurrent and attractor dynamics, oscillations and synchronization, and the large-scale architecture of the cortex. The book then turns to systems neuroscience: sensory processing in the visual and auditory systems, motor control, decision-making, memory, and the hippocampus.
The later chapters cover the deep connections between computational neuroscience and AI: what artificial neural networks borrowed from biology, what they did not, what modern deep learning has taught us about the brain, predictive coding, free-energy and Bayesian brain theories, and the role of neuromorphic hardware and spiking networks. The final chapters cover the frontier: connectomics, large-scale brain simulation, brain-computer interfaces, neurodegenerative disease, and the open questions that will define the next decade of the field.
Every chapter closes with a takeaway, so readers leave with both rigorous understanding of how biological circuits compute and a clearer view of what neuroscience and AI can actually learn from each other.
Autoren/Hrsg.
Weitere Infos & Material
The Brain as a Computer
It is Ari Chen's first week at Cortex Lab. On a Tuesday afternoon, he wanders into the main seminar room and finds Dr. Elena Vasquez already there, marker in hand, drawing on the whiteboard. She has divided the board cleanly down the middle. On the left, she is sketching a deep learning architecture: stacked rectangular layers connected by dense arrows, a loss function waiting at the bottom. On the right, she is drawing something Ari has never seen in quite this form: a cortical column, a vertical slice of brain tissue six layers deep, with neurons of different sizes and shapes arranged in bands, communicating up and down and sideways through a web of connections that looks nothing like the tidy parallel arrows on the left.
Elena steps back, looks at both drawings, and caps her marker. She points to the left diagram. "Every design decision in that network," she says, "was made by an engineer who had never seriously looked at this." She turns and points to the cortical column on the right. "That's the gap we're here to close." Ari stares at the right-hand diagram. He immediately notices that the cortical column has no obvious input or output layers, unlike the deep network. The connections loop back on themselves. There are inhibitory and excitatory cells, drawn with different symbols. The whole thing looks less like a stack and more like a living argument. He says nothing for a moment. Then he asks the only honest question available to him: "Where do I even start?"
Elena smiles. She pulls up a chair, sits down facing the whiteboard, and says: "Right here. You start with the question every computational neuroscientist asks before anything else: what problem is the brain actually solving?" That question is the starting point for this book. It shapes every chapter, every analogy, and every bridge we will build between biological and artificial intelligence. The whiteboard is the map. This chapter is where we learn to read it.
Diagram 1.1 - Two Diagrams on the Whiteboard
Why It Matters
If you work in machine learning or AI engineering, you already know how to build systems that learn from data. You understand gradient descent, attention mechanisms, and loss surfaces. You have shipped models. The question this book asks is not whether you can build AI systems. You clearly can. The question is whether the way you are building them is leaving an enormous amount of performance, efficiency, and capability on the table by ignoring the only existing proof of general intelligence we have ever found. That proof is the three-pound organ sitting inside every human skull.
Computational neuroscience is the discipline that treats the brain as an engineering puzzle. Not a medical puzzle, not a philosophical puzzle, but an engineering puzzle: given a physical substrate made of biological tissue, how does it process information, store it, retrieve it, use it to make decisions, and update itself based on experience? Those are engineering questions. They have quantitative answers. And the field that has been working on those answers for the past fifty years has accumulated a body of knowledge that is directly relevant to anyone building intelligent systems, and understanding that knowledge is no longer optional background reading for the serious AI practitioner. It is increasingly the map of where the next decade's ideas will come from.
This chapter lays the foundation. We will define what computational neuroscience is, how it differs from traditional neuroscience, and why the timing for engaging with it has never been better. We will meet the brain's basic statistics and let them settle. We will introduce the philosophical framework that organizes the field's questions. And we will preview the twelve-chapter journey ahead, so you can see exactly what you are committing to and why the sequence matters.
What Computational Neuroscience Is
People sometimes assume that computational neuroscience means using computers to study the brain: running simulations, processing brain-scan data, and applying machine learning to neural recordings. Those activities do happen inside the field, but they are not what define it. Computational neuroscience is defined by its central question, not its tools. The central question is: how does the nervous system process information? The word "process" is doing real work there. It means transforming inputs into outputs in a way that is useful to the organism. It means representing the world internally to make decisions. It means learning from experience and updating those representations over time. These are information-theoretic questions, and answering them requires building formal models of biological mechanisms rather than merely describing them.
The field grew out of a convergence of neuroscience, mathematics, and engineering that began in earnest in the 1980s. Before that convergence, most neuroscience was purely descriptive: cataloging cell types, mapping anatomical connections, and measuring electrical signals. All of that work was necessary and important. But description alone does not tell you how the system computes. For that, you need models.
A model, in the computational neuroscience sense, is a precise mathematical or algorithmic account of how a neural system transforms inputs into outputs. The model makes predictions. Those predictions can be tested against biological data. When the predictions fail, the model is revised. This cycle of model-build, test, and revise is the core practice of the field. It is the same loop that engineers use when designing any complex system, which is part of why engineers find the field accessible once they get past the biological vocabulary. The vocabulary is different. The epistemological stance is familiar.
Diagram 1.2 - What Computational Neuroscience Studies
Computational neuroscience works at every level of description simultaneously. At the smallest scale, it asks how a single neuron integrates thousands of incoming signals and decides whether to fire. At the next level up, it asks how small circuits implement operations such as edge detection or gain control. Further up, it asks how large networks generate attention, memory, and decision-making. At the largest scale, it asks how coordinated whole-brain activity gives rise to perception, action, and cognition. Each level has its own modeling tools, but all of them share one commitment: to describe what the system does in terms precise enough to generate testable predictions.
For someone coming from AI and machine learning, the most important thing to absorb early is that computational neuroscience is not reverse-engineering the brain to copy it wholesale. It is reverse-engineering the brain to understand what it is doing, which turns out to be a very different objective. Understanding what the brain is doing gives you principles to follow. Principles are portable. You can take a principle about how the visual cortex suppresses redundant information and ask whether the same principle could improve a convolutional network. Sometimes the answer is yes. Sometimes the biology does it in a way that has no clean analog in current architectures, which points you toward something genuinely new. That is the scientific frontier the book is navigating.
Neuroscience and Computational Neuroscience
The two fields share a subject matter: the nervous system. But they ask fundamentally different questions about it, and that difference matters enormously for how you should think about the knowledge each one produces. Elena uses a chess analogy she has been refining for years, and it is worth presenting here because it cleanly captures the distinction.
Imagine you want to understand chess. You could spend years cataloging every opening move and every pattern that has ever appeared on a board. You would learn an enormous amount. But you would not understand what a grandmaster is doing when she sacrifices a piece in the middle game to gain a tempo advantage fifteen moves later. That strategic understanding requires asking not just what the pieces can do, but what goals the player is pursuing, what trade-offs she is making, what model of the board's future she carries in her head.
Traditional neuroscience is largely in the business of describing the rules of the game: what cell types exist, how they connect, what chemicals they release, what electrical signals they produce. This is indispensable work. Without the rules, the strategic analysis is groundless. But computational neuroscience is in the business of explaining what the grandmaster is doing. It asks: given these physical pieces and these rules of connectivity, what computation is being performed, and why is that computation the right one for the task the organism faces? The two questions are complementary. The experimental neuroscientist and the computational neuroscientist need each other. The experimental data constrain the models; the models generate predictions that motivate the experiments. The field advances fastest when both operate in close collaboration, which is exactly the structure Cortex Lab was built to support.
Diagram 1.3 - Neuroscience vs Computational Neuroscience
A concrete example helps. Researchers in the 1950s and 1960s spent years carefully recording the electrical activity of neurons in the visual cortex of cats and monkeys, work pioneered by David Hubel and Torsten Wiesel. They found that individual neurons responded selectively to edges at specific orientations: a cell that fired vigorously when a bar of light was tilted at 45 degrees would barely respond to the same bar at 90 degrees. This was a remarkable and genuinely important discovery. But describing the selectivity is not the same as explaining it. Computational neuroscientists took that...




