Neural networks (NN) are algorithms used to detect information and conclusions from large sets of data by recognizing underlying relationships in sets of data the same way a human brain does. They hold a tremendous amount of potential for deep learning, part of a broader family of machine learning (ML) methods based on learning data representations, as opposed to task-specific algorithms. NN and deep learning are now computationally feasible due to GPUs, it shows unbeatable power on complex prediction problems that have very high dimensionality and millions-billions of samples.
However, for smaller scale problems this approach would likely be an overkill. Designing, building a NN, tuning hyperparameters, etc., requires specific experience and skills and time. In many cases it can not be justified even by a gain in model’s performance. Besides, it is not correct to juxtapose the two, because NN are part of ML. NN are designed to solve particular classes of problems, they are not universal solvers. In some cases even ML is overkill and a statistical test could be sufficient, for example.
Let’s say that you have a problem to model in order to make accurate predictions. We always kind of have an intuitive sense of the *variables/parameters* of this model. Learning is all about fitting a set of parameters to some data such that it can reproduce the data accurately. From traditional statistics to machine learning to deep learning, that’s always what’s being done.
How many parameters do you think would be able to model such a probability? A lot? A few? It’s a trivial example, but the same applies for less trivial ones where we know the numbers of parameters is limited, e.g. predicting the maximum speed of a car (weight, engine type, fuel, etc.). For these problem, it clearly looks like statistics would be able to fit the data. It seems linear, and relying on a few parameters only. Now, you want to classify emails whether or not they’re spam. How many parameters? Some words can help defining this is spam, sure, but do you have a full list?
I can also send spam by using a very standard language, without forcing you to buy or putting many links. It looks like the model will be richer than simply a few parameters. In fact, it looks like machine learning would help to learn a few hundred parameters to really be able to model what is a spam email.
Now you want to recognize faces on pictures. How complex do you think it is? Do all eyes look the same? Mouth? Skin color? Hair shape and color? Elements like eyes, nose and mouth only make sense when they are put together in a meaningful way (unlike Picasso likes to paint them). See, it looks like while our brains do it very easily, it’s really hard to grasp what makes a face. It’s a very complex and rich model that will be needed here, able to model A LOT of potential faces to efficiently recognize them.
Therefore, deep learning is most likely to yield better results than the previous approaches. Finally, if you select an approach too big for the task, you will simply 1) overfit the data and 2) spend WAY too much time building a complex approach for no benefit, quite the contrary…
“…And it is of course not true that we have to follow the truth. Human life is guided by many ideas. Truth is one of them. Freedom and mental independence are others. If Truth, as conceived by some ideologists, conflicts with freedom, then we have a choice. We may abandon freedom. But we may also abandon Truth.” – “How to Defend Society Against Science”, Paul Feyerabend
Paul Feyerabend (1924-1994), having studied science at the University of Vienna, moved into philosophy for his doctoral thesis, made a name for himself both as an expositor and (later) as a critic of Karl Popper’s “critical rationalism”, and went on to become one of the twentieth century’s most famous philosophers of science. An imaginative maverick, he became a critic of philosophy of science itself, particularly of “rationalist” attempts to lay down or discover rules of scientific method.
Hilma af Klint, Group X, No. 1, Altarpiece, 1915
Born to the son of a civil servant and a seamstress, Feyerabend took up reading as well as singing during his childhood. Having passed his final high school exams in March 1942, he was drafted into the Arbeitsdienst (the work service introduced by the Nazis), and sent for basic training in Pirmasens, Germany. Feyerabend opted to stay in Germany to keep out of the way of the fighting, but subsequently asked to be sent to where the fighting was, having become bored with cleaning the barracks! He even considered joining the SS, for aesthetic reasons. His unit was then posted at Quelerne en Bas, near Brest, in Brittany. Still, the events of the war did not register. In November 1942, he returned home to Vienna, but left before Christmas to join the Wehrmacht’s Pioneer Corps. Their training took place in Krems, near Vienna. Feyerabend soon volunteered for officers’ school, not because of an urge for leadership, but out of a wish to survive, his intention being to use officers’ school as a way to avoid front-line fighting. The trainees were sent to Yugoslavia. In Vukovar, during July 1943, he learnt of his mother’s suicide, but was absolutely unmoved, and obviously shocked his fellow officers by displaying no feeling.
In 1945, Feyerabend was shot in the hand and in the belly during the retreat from the Russian Army. The bullet damaged his spinal nerves. Two years later, he’d return to Vienna to study history and sociology at the University until later transferring to physics. His first article, on the concept of illustration in modern physics, published. Feyerabend could be described as “a raving positivist” at the time. The student found himself persuaded him of the cogency of realism about the “external world” (Popper’s important arguments for realism came somewhat later). The considerations Hollitscher deployed were, first, that scientific research was conducted on the assumption of realism, and could not be otherwise conducted, and, second, that realism is fruitful and productive of scientific progress, whereas positivism was simply a commentary on scientific results, barren in itself.
Feyerabend received his doctorate in philosophy for his thesis on “basic statements” in 1951. He applied for a British Council scholarship to study under Wittgenstein at Cambridge, but Wittgenstein died before Feyerabend arrived in England, so Feyerabend chose Popper as his supervisor instead.
In 1975, Feyerabend published his first book, Against Method, setting out “epistemological anarchism”, whose main thesis was that there is no such thing as the scientific method. Great scientists are methodological opportunists who use any moves that come to hand, even if they thereby violate canons of empiricist methodology.
In his article, “How to Defend Society Against Science”, the philosopher sought to defend society and its inhabitants from all ideologies, science included. All ideologies must be seen in perspective. One must not take them too seriously. One must read them like fairytales which have lots of interesting things to say but which also contain wicked lies, or like ethical prescriptions which may be useful rules of thumb but which are deadly when followed to the letter.
Now, is this not a strange and ridiculous attitude? Science, surely, was always in the forefront of the fight against authoritarianism and superstition. It is to science that we owe our increased intellectual freedom vis-a-vis religious beliefs; it is to science that we owe the liberation of mankind from ancient and rigid forms of thought. Today these forms of thought are nothing but bad dreams-and this we learned from science. Science and enlightenment are one and the same thing-even the most radical critics of society believe this. Kropotkin wants to overthrow all traditional institutions and forms of belief, with the exception of science. Ibsen criticizes the most intimate ramifications of nineteenth-century bourgeois ideology, but he leaves science untouched. Levi-Strauss has made us realize that Western Thought is not the lonely peak of human achievement it was once believed to be, but he excludes science from his relativization of ideologies. Marx and Engels were convinced that science would aid the workers in their quest for mental and social liberation. Are all these people deceived? Are they all mistaken about the role of science? Are they all the victims of a chimaera?
To these questions my answer is a firm Yes and No.
Now, let me explain my answer.
The explanation consists of two parts, one more general, one more specific.
The general explanation is simple. Any ideology that breaks the hold a comprehensive system of thought has on the minds of men contributes to the liberation of man. Any ideology that makes man question inherited beliefs is an aid to enlightenment. A truth that reigns without checks and balances is a tyrant who must be overthrown, and any falsehood that can aid us in the over throw of this tyrant is to be welcomed. It follows that seventeenth- and eighteenth-century science indeed was an instrument of liberation and enlightenment. It does not follow that science is bound to remain such an instrument. There is nothing inherent in science or in any other ideology that makes it essentially liberating. Ideologies can deteriorate and become stupid religions. Look at Marxism. And that the science of today is very different from the science of 1650 is evident at the most superficial glance.
For example, consider the role science now plays in education. Scientific “facts”are taught at a very early age and in the very same manner in which religious “facts”were taught only a century ago. There is no attempt to waken the critical abilities of the pupil so that he may be able to see things in perspective. At the universities the situation is even worse, for indoctrination is here carried out in a much more systematic manner. Criticism is not entirely absent. Society, for example, and its institutions, are criticized most severely and often most unfairly and this already at the elementary school level. But science is excepted from the criticism. In society at large the judgement of the scientist is received with the same reverence as the judgement of bishops and cardinals was accepted not too long ago. The move towards “demythologization,” for example, is largely motivated by the wish to avoid any clash between Christianity and scientific ideas. If such a clash occurs, then science is certainly right and Christianity wrong. Pursue this investigation further and you will see that science has now become as oppressive as the ideologies it had once to fight. Do not be misled by the fact that today hardly anyone gets killed for joining a scientific heresy. This has nothing to do with science. It has something to do with the general quality of our civilization. Heretics in science are still made to suffer from the most severe sanctions this relatively tolerant civilization has to offer.
Wolfgang Tillmans, Philharmonie Bloch III, 2017.
Is this unfair? Have I not presented the matter in a very distorted light by using tendentious and distorting terminology? Must we not describe the situation in a very different way? I have said that science has become rigid, that it has ceased to be an instrument of change and liberation, without adding that it has found the truth, or a large part thereof. Considering this additional fact we realize, so the objection goes, that the rigidity of science is not due to human will. It lies in the nature of things. For once we have discovered the truth. What else can we do but follow it?
This trite reply is anything but original. It is used whenever an ideology wants to reinforce the faith of its followers. “Truth” is such a nicely neutral word. Nobody would deny that it is commendable to speak the truth and wicked to tell lies. Nobody would deny that_-and yet nobody knows what such an attitude amounts to. So it is easy to twist matters and to change allegiance to truth in one’s everyday affairs into allegiance to the Truth of an ideology which is nothing but the dogmatic defense of that ideology. And it is of course not true that we have to follow the truth. Human life is guided by many ideas. Truth is one of them. Freedom and mental independence are others. If Truth, as conceived by some ideologists, conflicts with freedom, then we have a choice. We may abandon freedom. But we may also abandon Truth. (Alternatively, we may adopt a more sophisticated idea of truth that no longer contradicts freedom; that was Hegel’s solution.) My criticism of modern science is that it inhibits freedom of thought. If the reason is that it has found the truth and now follows it, then I would say that there are better things than first finding, and then following such a monster.
Tunga, Untitled, 2011, ink on paper, 29 7⁄8 × 20″. From the series “La voie humide,” 2011–16.
In approaching the topic of symmetry (in its many forms through nature, philosophy, music, and even logic), we find many different expressions of beauty. Symmetry itself becomes a feature that almost defines beauty in the way we can craft elegant equations in mathematics and physics to our own perceptions of facial features. In symmetry, we find a similarity among all these myriad forms of beauty, and, within symmetry itself, the repetition of a feature creates a sort of rhythm that invokes aesthetic pleasure. In searching for unifying principles among several different perceptions, subjective experiences, and even more objective forms of reasoning, we can view this sort of unity as something that creates defined, certain meaning among many forms. Symmetry becomes a rhythm, like the equality on both sides of an equals sign in a mathematical equation. And, in creating these uniformities among observations, judgements, and perceptions we can deepen our senses of the world and create discoveries in science and philosophy that we couldn’t have done before. Unity would seem to be a moment’s reflection will show us that unity cannot be absolute and be a form; a form is an aggregation, it must have elements, and the manner in which the elements are combined constitutes the character of the form. A perfectly simple perception, in which there was no consciousness of the distinction and relation of parts, would not be a perception of form; it would be a sensation. This sensation is the key to understanding the relation between moral value and aesthetic pleasure that the arts and sciences invoke within us.
Beauty in all forms, as aesthetic philosophers may pronounce, invoke physiological sensations with ourselves. Knowing and determining the nature of these sensations through our appreciation of art (and other aesthetic pleasures). We create them in ways we observe everyday and in anything. A pixel on a computer screen, the beat of a percussive instrument during a song, or even a vibration that travels through space and approaches our ears create patterns as they aggregate, combine, and form with one another. Whatever bodily change or effect of a nervous process that we experience as a result of that is our bodies method of interpreting and analyzing these aesthetic forms. Those who pay close attention to these sensations of their bodies and use that to discover new meaning, purpose, value, and other forms of wisdom can reap the benefits of these methods of reflection. But only through this close, careful introspection and reflection upon meaning and value through these aesthetic means (not only symmetry but other methods as well) can we begin to understand the nature of beauty. The form, brought upon by art and, especially symmetry, makes us more aware and sensitive to thought, ideas, principles, and means of imparting knowledge in making us human.
The part of aesthetic nature we find appealing is beauty of form. In this sense, form is these objects of beauty are expressed. In aesthetic terms, the rudimentary nature of formless stimulation is removed and from the emotional looseness of remaining lost in senseless thought.
Borrowing from the work of George Santayana, I believe there can be a coming together of beauty and form that a human being performs in the mind. We create inferences and insights about what we observe aesthetically and sense the unity as discussed earlier. Beyond the sensation itself and deep within the insights offered by works of art, we can detect the elements that underly beauty.
In Hiking with Nietzsche: On Becoming Who You Are, American professor of philosophy John Kaag shows how important and salient philosophy’s role in everyday life is. By hiking through mountains and experiencing what the Swiss Alps have to offer, Kaag illustrates a view of Nietzsche’s life that provides an intimate understanding of the challenges for which the German philosopher sought answers. Comparing himself to Zarathustra and Dionysus, Nietzsche actualizes his true potential in a way that other philosophers struggle with. He’s overcomes the limits and disadvantages of discourse and rumination and, instead, writes about the urgency of addressing issues of his time – many of which persist in the present day.
“As it turns out, to ‘become who you are’ is not about finding a ‘who’ you have always been looking for. It is not about separating ‘you’ off from everything else. And it is not about existing as you truly ‘are’ for all time. The self does not lie passively in wait for us to discover it.” I was incredibly satisfied by the immense level of reflection and thought put forward in analyzing and taking apart these arguments. It was a way of treating our thoughts and ideas as truthfully and justifiably as possible while still leaving room for the reader to maintain their own view of the issues Nietzsche brought up. It takes a tremendous amount of courage to address these issues, and, without the persistent and relentless work of both Nietzsche and Kaag, I’d struggle to even put these issues in words. I found the experience of reading the book absolutely insightful and eye-opening not only in the way Kaag depicted Nietzsche and the struggles he faced, but the way I related to them within myself. As I studied science and philosophy at Indiana University-Bloomington during my undergraduate years, I faced a tremendous amount of psychological and existential struggles. Things would get so worse with my mental health, social situations, academic performance, and even the thoughts I had about myself that, throughout senior year, I was just trying to leave my university as quickly and shamelessly as possible. I lost sight of the purpose in everything. My courses became tremendously more difficult, and I couldn’t ever figure out what to do. This book provides me with the ideals and arguments by which I can address those issues with far greater precision and clarity. I look forward to reading more Nietzsche in this thought-provoking and self-healing way such that I can continue to address these issues wherever I find them.
Even the pain and suffering that the individual experiences in society have home in Nietzsche’s work. By this, I mean that the way we react and deal with conflicts that cause us to suffer are taken with serious inquiry such that the individual can discover the true causes of what they are and the best ways to address them. As I read the book, I couldn’t help but compare the work and methods of the philosophers to how therapists approach individuals suffering from existential crises. A patient seeking help from an educated, wise therapist will often find him/herself at a loss of words and dumbfounded in terms of how to address his/her issues. It leaves the soul to suffer at the hands of a world that is wrathful, intimidating, and merciless. In concrete terms, this may include mental health issues such as depression or anxiety but also severe physical ailments such as cancer. Medicine and doctors should adhere to these truths and wrestle with them in their work in ways to treat patients and make the world a safer, healthier place by all means of measurement. Amazing work by physicians such as Richard Gunderman, Rita Charon, and Atul Gawande all hold the potential for making these changes happen. The way we internalize our suffering as part of a greater understanding of suffering that society has given us can let us internalize the reality of how and why we are meant to suffer. What might seem pessimistic and gloomy in our methods to understand the world turns out more encouraging and resilient to face whatever issues we experience in life.
Among the several lessons that Kaag and Nietzsche share discourse over include Nietzsche’s argument that self-discovery requires and undoing of the self-knowledge you assume you already have. This means that becoming yourself is a constant cycle between finding the self and also losing all sight of it. We can only truly become who we are as we overturn the fundamental truths and ideals that we believe make us who we are. This means there should be a level of trust and security as we perform these actions and do these things in life to become who we are. Nietzsche also elaborates that modern life distracts and deadens us in ways that prevent us from becoming who we are. The pleasures and fleeting desires of this world are nothing to compared to the near-unsurmountable challenge that is becoming who you are.
Kaag provides a clear example of these statements: “I remember too vividly an argument with my ex-wife that terminated with three words that I screamed before slamming our front door: ‘Let. Me. Be!’ I now know what I actually meant: ‘Get out of my way.’ Let me find my immutable essence. Unfortunately, there is no such thing as an immutable essence, at least not in my world. And so I left, but I never found what I was looking for, not even with (my new family) Carol and Becca. I found something else.” Carol is Kaag’s wife and Becca is his daughter.
Kaag can mention life story lessons as he ventures with his wife and daughter, and he draws upon his own personal experience in describing what Nietzsche himself sought to describe. The decadence, or decay, of the society around him, as Nietzsche noted, provides a careful, yet effective way of internalizing and dealing with the existential woes of today. As any philosopher dabbing in existentialism might come to realize, these concrete, realistic situations of philosophical truths come together in a neatly woven story. And the power which Nietzsche provided for his arguments has allowed them to resonate for decades.
As Nietzsche himself, said “It is an excellent thing to express a thing consecutively in two ways, and thus provide it with a right and a left foot. Truth can stand indeed on one leg, but with two she will walk and complete her journey.” (The Wanderer and His Shadow, 1880
Near the end of the book, Kaag explains how “‘Become what you are’ has been described as ‘the most haunting of Nietzsche’s haunting aphorisms.’” Indeed, it’s troubling to hear how who we are is something which we have to become, but that the thesis what we need to be is ourselves is all the more encouraging and reassuring for the reader.
In my current research on the zebrafish brain, I’m creating a mapping of parts of the brain to the genes which are expressed using mathematics and statistics. This method of devising theoretical models carries difficulties and issues in the way the accuracy and precision of these models. This model of the zebrafish neuroscience holds insight for our methods of using the organism for studying psychiatric disorders. In understanding phenomena of the brain, neuroscientists have various methods of referring to how to both explain and describe the causal mechanisms of the brain. The way our brain interacts with things like stimuli (such as visual imagery or sounds) and creates its own effects (such as neuronal responses in the brain) need to be precise to determine the nature of those phenomena we empirically observe. The 3M (model-mechanist-mapping) constraint is one such method.
In this post I will show plausibility that satisfying the 3M constraint gives us predictive, explanatory power in neuroscience that can be extended to cognitive science, psychology, and (pose the question for) consciousness. I’ll use various examples of neuroscience in proving its predictive power. I’ll also like to relate this predictive power to, at the very least, a basic form of consciousness. I hope to elucidate current findings in both science and philosophy as they relate to consciousness itself. We can begin this sort of inquiry with an overview of these neuroscientific explanations, then proceed to basic questions of how neuroscience relates to consciousness and what sort of empirical evidence has been shown towards this problem. Finally, we conclude with what limits scientists and philosophers currently face, and what anyone can do to meet those problems.
Alchemical Illustration from the Emerald Tablet of Hermes.
The Tablet had such an impact on the minds of histories greatest philosophers, esotericists and mystical thinkers, that it became the esoteric industry standard for every medieval and later renaissance system of alchemy.
The 3M has two claims. The first is that the variables in the model correspond to identifiable components, activities and organizational features that produces maintains or underlie the phenomena. The second is that the mathematical dependencies that are posited among the these perhaps mathematical variables within that model correspond to causal relations among the components of that mechanism. This mechanism-model-mapping (3M) constraint embodies widely held commitments about the requirements on mechanistic explanations and provides more precision about those commitments.
3M is much more than imposing an arbitrary rule on scientific theory, as David Kaplan, Lecturer in the Department of Cognitive Science at Washington University in St. Louis, explains. The demand follows from the many limitations of how predictions are formed and the conspicuous absence of an alternative model of explanation that satisfies scientific-commonsense judgments about the adequacy of explanations and does not ultimately collapse into the mechanistic alternative. The idea of being in compliance with the 3M constraint is shown to have considerable utility for understanding the explanatory force of models in computational neuroscience, and for distinguishing models that explain from those merely playing descriptive and/or predictive roles. Conceiving computational explanation in neuroscience as a species of mechanistic explanation also serves to highlight and clarify the pattern of model refinement and elaboration undertaken by computational neuroscientists. Under 3M, we can generally believe that the more accurate and detailed models are for target systems, the greater effectiveness they explain the phenomena.
One of the biggest setbacks of machine learning, as I’ve explained, is that models are far too descriptive of sets of data, yet not explanatory that they can be used for prediction. Scientist and philosophers debate whether 3M can explain phenomena in addition to describing them. I believe that dynamical and mathematical models in systems and cognitive neuroscience can generally explain a phenomenon only if there is a plausible mapping between elements in the model and elements in the mechanism for the phenomenon. In 1983 Professor of Psychology Philip Johnson-Laird expressed what was then a mainstream perspective on computational explanation in cognitive science: “The mind can be studied independently from the brain.” The extent to which this is true (which we call computational chauvinism, as did Piccinini in 2006) can be confirmed with our theoretical models of genetic mapping in the brain. However, we can argue forms of this computational chauvinism hold true as we bridge the gap between computational explanations and cognitive science. Our human cognitive capacities can be confirmed independently of how they are implemented in the brain. Delineating this computational chauvinism and predictive power of the 3M model, neuroscientists can have more power in their explanations of the brain.
Computational chauvinism is three claims: (1) computational explanation in psychology is independent from neuroscience, (2) computational notions are uniquely appropriate to psychological theory, and (3) computational explanations of cognitive capacities in psychology embody a distinct form of explanation. The neuroscientific and biological explanations and mechanistic explanations are covered by this form of explanation. These neuroscientific forms of explanation should prove insightful to the the two questions of consciousness, as explained by philosopher David Chalmers: generic and specific. Generic consciousness relies on the question of how neural properties explain the conscious state and the specific form, how they explain the content of the conscious state itself. To show that a computational analysis of neuroscience is possible, especially in the realm of consciousness, we need to refute the challenge fo computational chauvinism.
Computational chauvinism shares connections with functionalism, once the dominant position among philosophers of the mind (Putnam 1960). Functionalism, that the way a mental state functions determines what makes the mental state what it is, can be used to support the conclusion to abandon neuroscientific data. Canadian philosopher
Zenon Pylyshyn also argues these connections between computational chauvinism and functionalism.This comes as a result of the functionalist belief that psychology can explain phenomena independently of neuroscientific evidence. Drawing the analogy that the brain is similar to a computer, we imagine the functions of the mind as similar to running software. The computationalist neuroscientists believe the brain can be modeled as a computer. That the psychological phenomena can proceed without respect to the neuroscience means the brain is only the hardware of the computer and nothing else. Cognitive science would be the software that emerges. With this computer analogy, the functionalist would argue that the finding neural and computational explanations would be mostly irrelevant to psychology and cognitive science. At best, they may play a minor role in extreme examples of brain physiology. I will argue to refute functionalism to show the potential for the explanatory power of computational neuroscience.
More difficulties arise in our notions of objectivity with consciousness. At best, we can only observe behavior that tracks consciousness. We must use introspective forms of reasoning and thinking relate these subjective experiences to objective ideas and models of consciousness while appropriately measuring a subjective responses of consciousness. If I were to continue to stand by the explanatory power of computational neuroscience, it should hold the potential for this gap between the subjective and objective. The breadth of neuroscience, as it covers all forms of studying the brain and nervous system the constituents that make them up, we can look at the physical and mechanistic properties of the cerebral cortex for evidence of perceptual consciousness. My previous work on stochastic models of the brain should serve as a worthy example of this with the sense of vision. Looking at the general state of empirical work, especially as it relates to vision, give us a starting point for describing this consciousness.
believe I can argue that the constraint of 3M on explanatory mechanistic models because it can create the difference between phenomenological and mechanistic models as well as distinguishing between the possibility and actuality of the models. Phenomenological models provide descriptions of phenomena, and, as philosopher Mario Bunge argues, they describe the behavior of a target system without any unobservable variables (similar to the hidden variables I’ve described with causal models). In computational neuroscience, descriptive models (that summarize data effectively) differ from mechanistic models (that explain how neuroscientific systems work).
I cite the 1999 textbook Spikes: Exploring Computational Neuroscience as a seminal book in the scientific theories of computational neuroscience. The book sought to measure signals and responses from the nervous system and analyze those spike trains that followed. It uses several examples such as Gaussian waveform patterns and variations of the Hodgkin-Huxley models of firing neuron potentials. The latter model uses mathematics and conductance to explain how action potentials can be fired from neurons. These scientists, winning the Nobel Prize in 1963, performed that was very closely related to Biophysicist Richard Fitzhugh’s work in the 1960’s. Fitzhugh reduced the Hodgkin-Huxley model so that it could be visualized in phase-space and, therefore, use all variables at once, and, from this, be used for more accurate detailed predictions. I also believe this work distinguished the qualitative features of neurons on the topological properties of their corresponding phase space.
Kaplan explains that the model’s predictive power is weak. While it may generate accurate predictions about the action potential in the axonal membrane of the squid giant axon (their experimental system) to within roughly ten percent of its experimentally measured value, the critical question is whether it explains. Despite these features of the Hodgkin-Huxley models, these equations don’t explain how voltage changes membrane conductance. Scientists and philosophers who wish to use the predictive power of models in neuroscience require models to reveal the causal structures responsible for the phenomena themselves. Still, the Hodgkin-Huxley equations continue to provide the inspiration for interesting mathematics and physics problems.
The electrical activity can be physically measured from neuron cells and the scientists needed a way of determining the “spikes” (as the title suggests) that result from the data. At the time the book was written, there were many many other features of neurons, neural networks and brains that one would need to understand as well, no question about that. But the book sought to explain the spike (or action potentials) timing with as much accuracy and precision as possible. As neurons fire and send signals, they produce an action potential that’s created by the difference in charge along the neuron. From, using mathematical descriptions such as Bayesian formalism, the authors argue how to make sense of the neural data using probabilistic approaches to explain how stimuli may be predicted. Sensory neurons govern vision and we can gauge information processing by observing the potential of these receptive fields. The various electrical properties discussed in the book, such as spike rates, local field potential, and blood oxygen level dependent signal (BOLD), especially from groups of neurons and how they relate to one another provide the basis for these explanations of consciousness. Though “Spikes” was published in 1999, even as far back as 1990 were the biologist Francis Crick and neuroscientist Christof Koch describing how groups of neurons functioned together. Though they can be quantified mathematically, the exact nature of how they together relate to consciousness is not completely understood.
However, these neural sensory systems (the groups of neurons, pathways, and the parts involved in perception) do have potential about the subject’s environment. From this information we can create neural representations, which are the ways neural activity form to correspond to represent external stimuli that we readily observe. The closeness of this relationship, though, is hotly debated. Philosopher Rosa Cao argued that neurons will have little or no access to semantic information about the world, for example. Cao has also raised questions of what sort of functional units arise in describing neural representation. A very simple example I put forward is that information (in this case, representation of the relevant aspect of the stimulus that causes a neural response) is carried through series of spike potentials in the brain. Certain models that have been created from these data include the Dehaene-Changeux model which has been shown to create a global workspace for consciousness. By this explanation, a state must be accessible to be considered a consciousness state. A system X accesses content from system Y if (and only if) X uses that content in its computations/processing. It must be “globally” accessible to multiple systems including long-term memory, motor, evaluational, attentional and perceptual systems (Dehaene, Kerszberg, & Changeux 1998; Dehaene & Naccache 2001; Dehaene et al. 2006). This is irrelevant of whether the access is phenomenal.
Though I can’t make any statements of 3M directly in its relation to models of consciousness, I believe scientists and philosophers should begin observing the 3M criteria in their studies of consciousness. Researchers of any kind can raise questions of the explanatory power of these methods of describing physiological phenomena. We need a deep, precise explanation for our theory as they relate to forming predictions. Then, we can venture into the domain of consciousness with much more insight than without. The debates among mechanistic, dynamic, and predictivist explanations among functional and structural taxonomies. For all its flaws and limitations, mechanistic models of the brain still provide beneficial results to the answers at the core issues in philosophy of neuroscience, including explanation, methodology, computation, and reduction.
Sources
Bunge, M. (1964). Phenomenological Theories. In (Ed.) M. Bunge, The Critical Approach: In Honor of Karl Popper. New York: Free Press
Cao, Rosa. (2012). “A Teleosemantic Approach to Information in the Brain”, Biology and Philosophy, 27(1): 49–71. doi:10.1007/s10539-011-9292-0 ––– (2014). “Signaling in the Brain: In Search of Functional Units”, Philosophy of Science, 81(5): 891–901. doi:10.1086/677688
“A Neuronal Model of a Global Workspace in Effortful Cognitive Tasks”, Proceedings of the National Academy of Sciences of the United States of America, 95(24): 14529–14534. doi:10.1073/pnas.95.24.14529
Chalmers, David J. (1995). “Facing up to the Problem of Consciousness”, Journal of Consciousness Studies, 2(3): 200–219.
Crick, Francis and Christof Koch. (1990). “Toward a Neurobiological Theory of Consciousness”, Seminars in the Neurosciences, 2: 263–275.
Dehaene, Stanislas and Jean-Pierre Changeux. (2011). “Experimental and Theoretical Approaches to Conscious Processing”, Neuron, 70(2): 200–227. doi:10.1016/j.neuron.2011.03.018
Dehaene, Stanislas, Jean-Pierre Changeux, Lionel Naccache, Jérôme Sackur, and Claire Sergent. (2006). “Conscious, Preconscious, and Subliminal Processing: A Testable Taxonomy”, Trends in Cognitive Sciences, 10(5): 204–211. doi:10.1016/j.tics.2006.03.007
Dehaene, Stanislas, Michel Kerszberg, and Jean-Pierre Changeux
Fitzhugh, R. (1960). Thresholds and plateaus in the Hodgkin-Huxley nerve equations. The Journal of General Physiology 43 (5), 867–896.
Fitzhugh, R. (1961). Impulses and physiological states in theoretical models of nerve membrane. Biophysical journal 1 (6), 445–466.
Haken, Hermann, J. A. Scott Kelso, and H. Bunz. (1985). “A Theoretical Model of Phase Transitions in Human Hand Movements.” Biological Cybernetics 51 (5): 347–56.
Johnson-Laird, P.N. (1983). Mental Models: Towards a Cognitive Science of Language, 47 Inference and Consciousness. New York: Cambridge University Press.
Kaplan, D. M. (2010) “The Explanatory Force of Dynamical and Mathematical Models in Neuroscience: A Mechanistic Perspective” Philosophy of Science Vol. 78, No. 4.
Piccinini, G. (2006). Computational explanation in neuroscience. Synthese 153:343–353.
—– (2007). Computing mechanisms. Philosophy of Science 74:501-526. Piccinini, G., and Craver, C.F. (forthcoming): Integrating psychology and neuroscience: functional analyses as mechanism sketches. Synthese.
Putnam, H. (1960). Minds and machines. Reprinted in Putnam, 1975, Mind, Language, and Reality. Cambridge: Cambridge University Press.
Pylyshyn, Z. W. (1984). Computation and Cognition. Cambridge, MA: MIT Press.
There are generally two types of machine learning. Supervised learning is where we have a labeled dataset. This means we already have data from which to develop models using algorithms such as Linear Regression, Logistic Regression, and others. With this model, we can make predictions like, given data on housing prices, what will the cost of a house with a given set of features be. Unsupervised learning, on the other hand, doesn’t have a labeled dataset. The model we create in this setting just needs to derive a pattern amongst the data. We do this with algorithms such as K Means Clustering, K Nearest Neighbors, etc. to solve problems like grouping a set of users according to their behavior in an online shopping portal. But what if we don’t have much data? What if we are dealing with a dynamic environment and the model needs to gather data and learn in real time? Enter reinforcement learning. In this post, I’ll take a look at the basics of what reinforcement learning is, how it works and some of its practical applications.
Reinforcement Learning through Super Mario
Comparison with other Machine Learning Techniques
Without reinforcement learning, there is no supervisor to tell you if you did right or wrong. If you did well, you get a reward, otherwise you would not. If you did terrible, you might even get a negative reward. Reinforcement learning adds in another dimension – time. It can be thought of being in between supervised and unsupervised learning. Whereas in supervised learning, we have labeled data and unsupervised learning we don’t, in reinforcement learning, we have time delayed labels, which we call rewards. RL has the concept of delayed rewards. The reward we just received may not be dependent on the last action we took. It is entirely possible that the reward came because of something we did 20 iterations ago. As you move through Super Mario, you’ll find instances where you hit a mystery box and keep moving forward and the mushroom also moves and finds you. It is the series of actions that started with Mario hitting the mystery box that resulted in him getting stronger after a certain time delay. The choice we make now affects the set of choices we have in the future. If we choose a different set of actions, we will be in a completely different state and the inputs to that state and where we can go from there differs. If Mario hit the mystery box but chose not to move forward when the mushroom began to move, he’ll miss the mushroom and he won’t get stronger. The agent is now in a different state than he would have been had he moved forward.
If you were playing Super Mario Bros. for the first time, you might have started with a clean slate – not knowing what to do. You see an environment in which you as Mario, the agent, have been placed that consists of bricks, coins, mystery boxes, pipes, sentient mushrooms called Goomba, and other elements. You begin taking actions in this environment by pressing a few keys before you realized then you can move Mario with the arrow keys to the left and right. Every action you take changes the state of Mario. You moved to the extreme left at the beginning but nothing happened so you started moving right. You tried jumping onto the mystery box after which you got a reward in the form of coins. Now, you learned that every time you see a mystery box, you can jump and earn coins. You continued moving right and then you collided with a Goomba after which you got a negative reward (also called a punishment) in the form of death. You could start all over again, but by now you’ve learned that you must not get too close to the Goomba; you should try something else. In other words, you have been “reinforced”. Next, you try to jump and go over the Goomba using the bricks but then you’d miss a reward from the mystery box. So you need to formulate a new policy, one that’ll give you the maximum benefit – gives you the reward and doesn’t get you killed. So you wait for the perfect moment to go under the bricks and jump over the Goomba. After many attempts, you take one such action that causes Mario to step over the Goomba and it gets killed. And then you have an ‘Aha’ moment; you’ve learned how to kill the threat and now you can also get your reward. You jump and this time, it’s not a coin, it’s a mushroom. You again go over the bricks and eat the mushroom. You get an even bigger reward; Mario’s stronger now. This is the whole idea of reinforcement learning. It is a goal-oriented algorithm, which learns techniques to maximize the chances of attaining the goal over many iterations. Using trial and error, reinforcement learning learns much like how humans do.
AlphaGo
Reinforcement learning hit the big time in March 2016 when DeepMind’s AlphaGo, trained using RL, defeated 18-time world champion Go player Lee Sedol 4-1. It turns out the game of Go was really hard to master for the machine, more so than games like Chess simply because there are just too many possible moves and too many numbers of states the game can be in.
Just like Mario, AlphaGo learned through trial and error, over many iterations. AlphaGo doesn’t know the best strategy, but it knows whether it won or lost. AlphaGo uses a tree search to check every possible move it can make and see which is better. On a 19×19 Go board, there are 361 possible moves. For each of these 361 moves, there are 359 possible second moves and so on. In all, there are about 4.67×10^385 possible moves; that’s way too much. Even with its advanced hardware, AlphaGo cannot try every single move there is. So, it uses another kind of tree search called the Monte Carlo Tree Search. In this search, only those moves that are most promising are tried out. Each time AlphaGo finishes a game, it updates the record of how many games each move won. After multiple iterations, AlphaGo has a rough idea of which moves maximizes its chance of winning.
AlphaGo first trained itself by imitating historic games played between real players. After this, it started playing against itself and after many iterations, it learned the best moves to win a Go match. Before playing against Lee Sedol, AlphaGo played against and defeated professional Go player Fan Hui 5-0 in 2015. At that moment, people didn’t consider it a big deal as AlphaGo hadn’t reached world champion level. But what they didn’t realize was AlphaGo was learning from humans while beating them. So by the time AlphaGo played against Lee Sedol, it had surpassed world champion level. AlphaGo played 60 online matches against top players and world champions and it won all 60. AlphaGo retired in 2017 while DeepMind continues AI research in other areas.
It’s all fun and games, but where can RL be actually useful? What are some of the real world application? One of the largest field of research and now beginning to show real promise is the field of Robotics. Teaching a robot to act similar to humans has been a major research area and also part of several sci-fi movies. With reinforcement learning, robots can learn similar to how humans do. Using this, industrial automation has been simplified. An example is Tesla’s factory that consists of more than 160 robots that do a large part of the work on cars to reduce the risk of defects.
RL can be used to reduce transit time for stocking and retrieving products in the warehouse for optimizing space utilization and warehouse operations. RL and optimization techniques can be utilized to assess the security of electric power systems and to enhance Microgrid performance. Adaptive learning methods are employed to develop control and protection schemes, which can effectively help to reduce transmission losses and CO2 emissions. Also, Google has used DeepMind’s RL technologies to significantly reduce the energy consumption in its own data centers.
AI researches at SalesForce used deep RL for automatically generating summaries from text based on content abstracted from some original text document. This demonstrated an approach for text mining solution for companies to unlock unstructured text. RL is also being used to allow dialog systems (chatbots) to learn from user interactions and help them improve over time. Pit.AI used RL for evaluating trading strategies. RL has immense applications in the stock market. Q-Learning algorithm can be used by anyone to potentially gain income without worrying about market price or risks involved. The algorithm is smart enough to take all these under considerations while making a trade.
A lot of machine learning libraries have been made available in recent times to help data scientists, but choosing a proper model or architecture can still be challenging. Several research groups have proposed using RL to simplify the process of designing neural network architectures. AutoML from Google uses RL to produce state-of-the-art machine-generated neural network architectures for language modeling and computer vision.
Meandering through information from different disciplines is difficult for anyone – be them a scientist, philosopher, or anything else. On his website and in this interview, we’ll take a look at how Adam Kruchten learned to figure out what guided him in his passions and how he applies both scientific and philosophical thinking to understanding statistics. HA: Adam, as an undergraduate, you studied mathematics and philosophy. Now you’re going to enroll at University of Pittsburgh to study biostatistics. How did you go from being interested in mathematics and philosophy to biostatistics? Adam: Statistics, and inference more generally, in some form has always been of interest to me, it just took me quite some time to learn that about myself. Early in my undergraduate career I worked in research in statistical mechanics, and I was always fascinated by the probabilistic models. Idealizations could capture tremendous amounts of useful information about extraordinarily complex phenomena. Further, the same underlying notions of probabilistic modeling could be used to understand and cope with both true randomness and epistemological uncertainties without any difference in mathematics. Originally I thought I was mostly drawn in by the physics. I realized later that the physics, while interesting, was not what drew me in. It was really the methodology. I hopped around different fields, but had the same problem. Eventually I settled on math and philosophy, and there I found fields where I could study and understand fundamental issues underlying robust scientific inferences. In math I was drawn to logic, and in philosophy I was drawn broadly to issues addressing philosophy of science: philosophy of science proper, but also language, epistemology, and metaphysics.
After graduation I took a job in applied mathematics, but my role was really mostly an applied statistician. Here I worked closely with a professor of statistics and found the underlying study of inferences that had really drawn me to numerous fields prior.
As for biostatistics specifically rather than statistics more generally? Biostatistics occupies its place inside public health programs. I think applying statistics to public health issues is a great way to make a meaningful impact through the study and application of my underlying passions.
HA: What role does (or will) philosophy play in your research? How do you hope to study science and philosophy hand in hand? Adam: Beyond philosophy directly informing my statistical work, I would also like to eventually research questions that are fundamental to inference itself. When doing this kind of research you are not just relying on philosophy, you are directly doing philosophy.
HA: On your blog you’ve written about the philosophical thesis of physicalism in a way that people without a strong background in philosophy can understand (https://adamkruchten.wordpress.com/2018/05/07/you-are-not-your-brain/). What sort of understanding do you think this general audience should have of philosophy? Adam: I try to write in an accessible way that doesn’t require much philosophical understanding, but I think I do expect readers to at the very least think “like a philosopher.” By “think like a philosopher,” I really mean several things. You should read with curiosity and openness: reading while prepared to dig deeper into elements you may not understand and with a willingness to change your own views as necessary. At the same time I think you should read with a critical but charitable mind. Critical, meaning you look for implicit assumptions, look for leaps in logic, and rigorously assess the foundations of any premises. Charitably, meaning you only attempt to criticize the best possible version of the argument: don’t set up straw men, see if small errors in argument and prose can be easily corrected, and engage with the mindset that an argument was made in good faith.
HA: A bit more specific, what can scientists do to appreciate philosophy better? Adam: There’s an obvious answer here which is just “read more philosophy.” This is an honest answer, but it only goes so far. I think reading more philosophy is always useful, but there is far more philosophy than even a professional philosopher could read and understand, let alone someone with a career outside of the field.
For a more practical answer I think scientists should engage in science the same way I answered the previous question. Think like a philosopher by acknowledging and assessing underlying premises and methodological assumptions in doing science. HA: Before we finish, what’s one book everyone should read? Adam: This is a tough one. I have a hard time suggesting one book for a variety of reasons. I think I will answer with the book I feel most influenced my thought, Immanuel Kant’s Prolegomena to any Future Metaphysics. This book is Kant’s own summary of the much longer Critique of Pure Reason. I think that reading this book shed a great deal of light on various ways of thinking I had taken for granted, and helped me come to terms with a lot of what I had, at times erroneously, assumed implicitly to be true about the world. Just as Hume awoke Kant from his dogmatic slumber, so did this book for me.
Neuroimaging, ways of understanding how the brain produces images, produces sets of data that are high-dimensional and complicated. Ways of interpreting this data provides the means for understanding how the brain encodes and decodes images. In this context, encoding refers to predicting the imaging data given external variables, such as stimuli descriptors and decoding refers to learning a model that predicts behavioral or phenotypic variables from fMRI data. With the way these models can be learned and predicted, supervised machine learning methods can be used to decode images to relate brain images to behavioral or clinical observations. Sci-kit learn can be used for this analysis in making predictions that can be cross-validated.
I’ve explored Nilearn, a Python module that uses simple interfaces for people to apply machine learning to neuroimaging data. This module would allow me to get the best visualizations for raw data and processed results, and it’s built on scikit-learn, a popular Python machine learning module.
In my fMRI project, I re-create the methods of Miyawaki et al. (2008) in inferring visual stimulus from brain activity. In the experiment of Miyawaki et al. (2008) several series of 10×10 binary images are presented to two subjects while activity on the visual cortex is recorded. In the original paper, the training set is composed of random images (where black and white pixels are balanced) while the testing set is composed of structured images containing geometric shapes (square, cross…) and letters. I will use the training set with cross-validation to get scores on unknown data. I can examine decoding (the reconstruction of visual stimuli from fMRI) and encoding (prediction of fMRI data from descriptors of visual stimuli). This would let me look at the relation between stimuli pixels and brains voxels from both angles. The approach uses a support vector classifier and logistic ridge regression in the prediction function in both the decoding and the encoding.
This June, I’ll begin work in a neuroscience lab where I will be using computational methods to study the zebrafish brain. I hope to cultivate more skills as part of this intrinsic, self-driven passion for neuroscience from a computational perspective. The dynamic interplay of experimental and theoretical models in evaluating and re-evaluating hypotheses is fascinating.
References:
Miyawaki, Y., Uchida, H., Yamashita, O., Sato, M.-A., Morito, Y., Tanabe, H. C., et al. (2008). Visual image reconstruction from human brain activity using a combination of multiscale local image decoders. Neuron 60, 915–929. doi: 10.1016/j.neuron.2008.11.004
I’m proud to announce the launch of SeqAcademy.org, a website to teach others RNA-Seq and ChIP-Seq analysis even with no prior programming experience.
This project was the result of the NIH April Data Science Hackathon, at which researchers from across the globe met to work on projects here at the NIH. Our group created SeqAcademy, an educational pipeline for RNA-Seq and ChIP-Seq designed to teach others the basics of bioinformatics. I hope to use this website to teach myself html/css, reach out to others, and provide for the greater bioinformatics community. I could see this project gaining momentum and becoming something greater – even with letting users login to the website, follow tutorials at their own pace, and earn certificates of completion. I would love to take SeqAcademy the step further and develop it into whatever the bioinformatics community needs as far as tutorials are concerned. It would require a lot of work creating a website from scratch, but it has the potential to help a lot of people.
“Everyone should have a deep understanding of science.” It seems like a lofty ideal. While it’s one thing for the general public to respect scientists for their work, it’s another to ask them to understand it on a deep level. As scientists and science writers share knowledge with others, we get a glimpse into their minds. Communicators like Neil deGrasse Tyson popularize astrophysics in such a way that the audience feels at ease with scientific jargon or conversations of the universe. In his new book Astrophysics for People in a Hurry, he promises this level of conversation for a non-scientific audience. Everyone develops a kind of understanding similar to theirs, and it’s more of a shared appreciation than a test of intelligence.
With his history of the universe, Tyson is off through space and time. In about 14 billion years, the expanding universe that began from the size of water droplet grew to today’s observable universe of 46 billion lightyears. Precision and detail are found sprinkled throughout Tyson’s story as he explains how the four fundamental forces of physics and phase changes of matter came about and interacted with one another. The reader feels comfortable with galaxies, planets, and dark energy with Tyson’s style of sharing how much time has passed and how much longer the reader will need to hold on. It feels as though the individual events unfold with respect to a greater purpose or narrative. Though the book is a set of essays, they’re presented like a conversation over tea with Tyson himself. Everything from Tyson’s background as a black astrophysicist to his religious (or lack thereof) convictions come about in this narrative.
Popular science is popular in some ways through awe. Stories that capture the public’s imagination – especially Tyson’s astrophysics tales – provide a public engagement that has not only instilled empathy in individuals but shaped policy on a larger scale. In astrophysics, the images from major telescopes like the Hubble and James Webb wouldn’t have been possible without the popular opinion swaying in their favor. Online science projects like Zooniverse and Foldit rely on crowd-sourced efforts of individuals to, respectively, volunteer projects and find protein structures. Everyone – scientists and non-scientists alike – becomes part of the same unified project this way. Greater purposes, narratives, and everyone’s place in the universe make sense on a different level through these projects. Like gladiators in a coliseum – the stories of science are shown to the spectators. As scientists and writers share the stories, everyone is intrigued in wonder. It’s exciting and thrilling to look at scientific phenomena in different ways – each one challenging everyone’s assumptions and ideas. Tyson’s book – and the rest of science communication – educate the public through these dimensions, and, while scientists keep speeding ahead through the universe, the rest of society can stand comfortably behind them knowing they’ll still catch up.
Heuristic 