10 ms·

Antifragility in complex dynamical systems

173 points by RafelMri 2y ago

bob1029 2y ago
I've found that intentionally causing abrupt (but reasonable) changes to hyperparameters in evolutionary spiking neural network simulations (I.e between each generation) results in far more robust simulations that will meet fitness criteria with less likelihood of getting stuck somewhere. The tradeoff being that simulating will take longer, but this may be worth things like reducing the chances of your resource requirements going asymptotic half way through. My current method of perturbation is to cap the total # of activations per candidate globally. When at least 3 improvements in the best global fitness score have been achieved, I cut the global activation limit in half. Obviously, this is fairly catastrophic for the current population. Each generation can increase the limit by 1% if no improvements made. This provides aggressive selection pressure for more efficient networks and forces a different kind of partial restart of the training process (I.e. do what you just did with 50% resources now). Very often this does result in optimization of the network. It also seems to make the networks rebound faster after restarts. At first, a restart can take 10-20 generations to recover. By generation 1000, this is closer to 1-2.
- _yb2s 2y ago
  Very cool, but I'm surprised you're sharing this on here and not in a job interview with a deep learning startup and/or an arXiv paper.
  - bob1029 2y ago
    > a job interview with a deep learning startup I am not aware of any DL startups that are interested in techniques which are incompatible with GPUs and back propagation. I think this is a solo journey through the dark forest for now.
    - inciampati 2y ago
      Eloquent, realistic, and sad. I'm living my own version of this (in biotech). The journey is the wonder.
    - phyalow 2y ago
      Correct me if im wrong, this is just a question originating from my own curiosity, but could you not apply the same or similar techniques to classical NN architectures? I dont see exactly how the differentiability of the architecture matters? I.e. in between epochs randomly perturb weight values (similar effect I understand to introducing scheduled noise or non-isomorphic transforms on the features), implement a scheduled causal drop in/out scheme etc? I know of people that train models these ways as a form of regularisation and they claim better fits too. Enjoy your adventures in the Dark Forest.
      - bob1029 2y ago
        > I dont see exactly how the differentiability of the architecture matters? I don't know that classical architectures are infeasible to perturb as noted in other comments. I just think we have a really interesting knob to turn in this specific case that doesn't seem to exist in other architectures. How would you go about constraining the resource consumption (by up to half!) of a classical ANN while preserving realistic hopes that it will recover between training epochs? And, we are talking real resource reduction here not some academic thing. You can see the difference in task manager. That's actually the primary tool I've been using to determine how to tune the parameters of this scheme.
        SuchAnonMuchWow 2y ago
        There has been some work to dynamically reduce the compute required by a network. See for example: https://arxiv.org/abs/2404.02258 https://arxiv.org/abs/2404.02258 They have a fixed compute budget which is lower than what the LLM need, and dynamically decide to allocate this compute budget to different part of the network. So its not exactly what you propose since here the compute budget is fixed (that's the point of the paper: to make the network learn how to allocate the resources by itself) but its dynamic for each part of the network, so it shows that its possible.
    - sva_ 2y ago
      > I think this is a solo journey through the dark forest for now. In academia there's quite a bit of stuff on neuromorphic computing going on.
  - sethammons 2y ago
    Your surprise breaks my brain. Why would someone "keep it to themselves" outside some white paper they won't have time to promote or for a job interview that they might not need? Is your thought that it is some novel approach that nobody is doing and has obvious market implications? Even then, the implications matter only if you are in a position to take advantage. The Internet used to be a place people shared cool shit because it was cool, not for some monetization scheme or personal gain beyond the joy of sharing. Content creators (as a job) and hustle culture are a blight.
    - baxtr 2y ago
      They are sharing it here?! It’s not like OP is keeping it to themselves
    - _yb2s 2y ago
      As I'm sure you're aware deep learning work better is the most hot and highly funded thing right now... there is nothing wrong with making sure you get the credit for a new technique, and it in no way precludes widespread free sharing of the idea. I am an academic researcher (not in this field) and everything I do is 100% free and public, but I still either patent or publish it, because getting the credit is what allows me to keep doing what I love. I actually hate that part, and would rather just do the work, but I already push that line too far sometimes, and risk losing my funding aka "publish or perish."
  - authorfly 2y ago
    Respect to the OP as spiking networks are a different domain, but the field of Graceful Degradation and Perturbations itself is almost older than the field of Neural Networks itself. Generally, the bitter lesson occurs - these techniques show some kind of useful robustness at parameter M, but fail to show as much of a useful robustness relatively as parameter M*100. For example, before attention, using Drop out (in essence a form of forcing robustness) was one of the most common techniques. The quality of robustness or its resulting sub parts (e.g. the tradeoff between precision and recall) is incorporated as a measure in some of the fields basic scores, such as harmonic F1 score. Still it, like alternate activation functions, data beyond matrices and vectors and non-linear networks, remain very unstudied.
- JackeJR 2y ago
  I am quite curious about your approach, Do you have an example code somewhere?
  - bob1029 2y ago
    I'd start here: https://arxiv.org/abs/2406.12552 https://arxiv.org/abs/2406.12552 I will say that less is more and "The Bitter Lesson" applies here. Chasing biologically-inspired rabbits, such as STDP and LIF (see paper above/wikipedia), does seem to be a waste of time, especially when we have this thing entirely outside of biology that can arbitrarily replicate, serialize, mutate and simulate billions of instances of the same parent candidate in minutes-hours. Leaky charge carriers and inability to persist learned weights between candidate instantiations are limitations, not features to emulate. Imagine if you could be reincarnated with all of the exact knowledge you have today. Then clone that 1000 times and apply subtle mutations to it. Then, put these clones in an FFA arena and see who wins based upon a very well crafted fitness function. Then, do all of that over and over thousands of times per hour.
    - jcims 2y ago
      >Chasing biologically-inspired rabbits, such as STDP and LIF (see paper above/wikipedia), does seem to be a waste of time Unless physiological compatibility with a biological brain is considered an interesting endpoint? If you think about Neuralink for example, wouldn’t it be interesting if our brains could directly engage the model? Not just a translation layer but perceive it directly and natively though some kind of synaptic modem that converts analog exchanges of neurotransmitters to the synthetic network in the digital domain.
      - bob1029 2y ago
        > Unless physiological compatibility with a biological brain is considered an interesting endpoint? Excellent point. I am focused entirely on synthetic simulations without biological integration (for now). I think that could be an interesting next step - determining a way to modify a SNN that was trained in a synthetic time domain such that it can process real time signals. Training these things online is not really an option. You have to run a large # of offline simulations before you can find the right networks. Learning rules like STDP could theoretically address the online learning problem, but I couldn't find any traction down that path yet.
- taylorius 2y ago
  Firstly, I'm coming at this from the pint of view of standard feed forward NNs, rather than spiking neural networks, so apologies in advance if I am way off-base. By global activation limit, I assume you mean the total sum of neuron activation during one forward pass? So halving it would strongly encourage sparsity? How do you enforce the reduced limit? Just put it in the loss function? Or does the spiking nature allow you to "sparsify" the spiking by raising a threshold? Very very interesting stuff. I'm working with differential evolution, and incorporate correlation measures between candidates to (try to) allow for multiple divergent solutions, and it also feels fairly off the beaten track, compared to mainstream AI. But who cares, let's enjoy the dark forest! :-)
  - bob1029 2y ago
    > By global activation limit, I assume you mean the total sum of neuron activation during one forward pass? Yes. For each candidate network I keep a counter of the total # of neuron activations. This figure is constrained by a global limit (which varies based on prior performance). We simply break out of the simulation loop once the allotted resources are exhauted. This means that more efficient networks are more likely to score higher.
- seanhunter 2y ago
  > I've found that intentionally causing abrupt (but reasonable) changes to hyperparameters in evolutionary spiking neural network simulations (I.e between each generation) results in far more robust simulations that will meet fitness criteria with less likelihood of getting stuck somewhere. If I understand you correctly this is similar to the philosophy behind simulated annealing isn't it? The idea being that adding some peturbation can be enough to jog the evolution out of local minima and you probably want to reduce the amount of those jogs over time so it ends up converging. https://en.wikipedia.org/wiki/Simulated_annealing https://en.wikipedia.org/wiki/Simulated_annealing
  - neom 2y ago
    Not op but what they described is indeed sim annealing coupled with relaxing feedback loops. This approach would in theory result in prevention of stagnation, forced adaptation and over time, creating some form of resiliency. Pretty cool stuff!
  - bob1029 2y ago
    I was actually thinking about simulated annealing as I got into this. I've found that reducing this specific mode of perturbation over time has not been necessary. The simulation will oscillate its way right into a perfect fitness score assuming everything else is good. Convergence seems to rely much more heavily on the quality of the fitness function and population management (i.e. size, selection criteria, mutation strategies). I apply random mutations to delays and weights, but using a spectrum of mutation each time (I.e. some clones are heavily mutated while others are barely touched). This spectra of mutation is constant all the way through the simulation.
- lo0dot0 2y ago
  Some of you may die... But that's a sacrifice i'm willing to make.
adiM 2y ago
Is this just risk seeking behavior (in contrast to risk sensitive)
- marci 2y ago
  Now I wonder what a machine learning model with risk seeking behaviour would look like.
  - mike_ivanov 2y ago
    https://news.ycombinator.com/item?id=41231644 https://news.ycombinator.com/item?id=41231644
    - marci 2y ago
      I had read this comment, but didn't read it under this lens. Thanks for (re)pointing me to it.
  - nottorp 2y ago
    From the little I know about ML in general, you always want to introduce randomness so you don't get stuck in a local minimum/maximum. So they're all risk seeking?
delichon 2y ago
Systems or organisms can be defined as antifragile if they derive benefit from systemic variability, volatility, randomness, or disorder. If that's a riddle, death fits. Those things are characteristic of the pestilence, famine, war, etc. that feed death. The decay surrounding death is complex and dynamical. And death is more sustainable than any living system.
- _yb2s 2y ago
  but death isn't a system or organism...
  - fsckboy 2y ago
    >but death isn't a system death is entropy. do you know of a system without entropy? >or organism... boy are you going to be surprised when you answer that knock on the door, eh?
    - BriggyDwiggs42 2y ago
      Isn’t it definitely not an organism? If death is the inevitable process from order to disorder, isn’t it deeply a-systemic at its core? If it were a system, it would also eventually have to end.
      - hobs 2y ago
        Arguably it does, the heat death of the universe option would spell the end of entropy, and the end of death (unless you believe in a conformal cyclic universe)
        BriggyDwiggs42 2y ago
        Actually yeah fair enough; I’ll grant that death may be a system.
        _yb2s 2y ago
AgentOrange1234 2y ago
I wish folks would write more plainly. “Antifragility characterizes the benefit of a dynamical system derived from the variability in environmental perturbations.” Geish.
- _yb2s 2y ago
  Well it is in a journal called "complexity." But I agree, that sentence is awful, I'm surprised they decided to lead with it, and nobody proof reading the paper objected.
  - defrost 2y ago
    The leading three authors are from Germany, Mexico, and Switzerland resptfully. The sentence they want should capture the notion of being robust when poked with a stick. A pencil on it's tip is a fragile system, one burp and it falls to the table, far from the intial state. Marbles in fruit bowls are anti-fragile, given a good shake (up to a threshold) and they remain in the bowl and return to the low centre. Reading further, they want more; that repeated perturbations should deliver benefit, that systems in an warped egg carton configuration can be annealed to reveal an optimal point by vigorous shaking slowly reduced in degree. Having now read the whole paper it doesn't seem novel wrt to the state of dynamic system discussions in the mid 1980s other than the insertion of "antifragility" as nomenclature into the mix.
    - wenc 2y ago
      > A pencil on it's tip is a fragile system, one burp and it falls to the table, far from the intial state. Marbles in fruit bowls are anti-fragile, given a good shake (up to a threshold) and they remain in the bowl and return to the low centre. I believe what you're describing is a stable vs unstable system, not fragile vs antifragile. You can perturb a bowl with a marble, and the marble will still end up in the middle because it will return to stable equilibrium point (an "attractor"). Yours is an illustration of a stable system. Whereas a marble placed in an upside down bowl (with no ridges, just a half sphere), when perturbed, will fall off. This is an unstable system. These are classic examples used in (Lyapunov) stability theory. Fragility and antifragility aren't about stability (returning to equilibria), but gains or losses after perturbation, which is related to convexity/concavity. When you perturb an anti-fragile (or convex) system, it doesn't return to equilibrium but in fact improves. Conversely, when you perturb a fragile system, it degrades. The analysis is usually done with Jensen's inequality rather than Lyapunov. EDIT: not sure why the downvotes. I'm pointing out a fact. The examples do not demonstrate antifragility, but stability, which is not the same concept.
_yb2s 2y ago
I love how Taleb managed to translate an ideal from stoic philosophy into a precisely defined mathematical concept
- Animats 2y ago
  Some of this reads like rediscovering control theory, which is all about stability and robustness within defined limits. It's more popularization than innovation. The person who first translated this idea into math was James Clerk Maxwell, in his paper, "On Governors", in 1868.[1] Maxwell was the first to get a mathematical handle on stability of feedback systems. He wrote: "If, by altering the adjustments of the machine, its governing power is continually increased, there is generally a lin1it at which the disturbance, instead of subsiding more rapidly, becomes an oscillating and jerking motion, increasing in violence till it reaches the limit of action of the governor. This takes place when the possible part of one of the impossible roots becomes positive. ... I have pointed out that, under certain conditions, the sudden disturbances of the machine do not act through the differential system on the governor, or vice versa. When these conditions are fulfilled, the equations of motion are not only simple, but the motion itself is not liable to disturbances depending on the mutual action of the machine and the governor." That, right there, is the beginning of the mathematics of antifragility. Today, we call this turning up the P term of a PID controller too high and getting outside the stable region. Maxwell was the first to formalize the role of delay and damping (he calls it "viscosity") in this. There's "robust control", a useful theory of "antifragility" in industrial use for decades. Here's an excellent video on how this is done.[2] That's worth a watch. Note the emphasis on how the combination of delay and noise can cause instability in a system resistant to both delay and noise separately. This can be dealt with by moving the areas of instability around in the frequency domain. (Modern control theory has become insanely complicated over the last few years. Until recent years, control theory was mostly linear control theory plus some self tuning. Now, control theorists are trying to figure out how to get the benefits of machine learning while still having predictable error bounds. This means getting into the innards of what neural nets are really doing. I still get IEEE Trans. on Control Systems Technology, but don't understand it any more. So I'm not sure how that's coming along.) [1] https://archive.org/details/jstor-112510/mode/2up?view=theater https://archive.org/details/jstor-112510/mode/2up?view=theat... [2] https://www.youtube.com/watch?v=A7wHSr6GRnc https://www.youtube.com/watch?v=A7wHSr6GRnc
krick 2y ago
There are now so many bullshit-terms derived from Taleb's bullshit books. Like how people still keep mentioning "black swans" as if it actually means something other (or something more) than "unexpected event". And for some unfathomable reason it keeps traction. Similarly how Mandelbrot "redefined" (i.e. distorted) the meaning of "Lindy effect", and it stuck (however, I didn't notice if it became popular to call a millenia-old banality by than name after Mandelbrot, or Taleb again). It probably should serve as an another example of that if you are arrogant enough, people will follow you just because of that. However, I guess "antifragility" isn't the worst of these.
- aaron695 2y ago
  [dead]
- nottorp 2y ago
  When I saw the title I thought it came out of one of those online bullshit generators...
- padolsey 2y ago
  I don't much mind the populurization of these terms or making people think in more novel ways. Even if it's seen as BS by those more informed, I'm glad such books are written and communicators like Taleb exist. Without him, I wouldn't have discovered a bunch of tangential things. I will admit, it gives my brain a satisfying itch too, as I realise that academia is often just the refined encoding of pretty mundane everyday truths, so when someone is able to come in and re-extract that and share it widely, even with a bit of gentle-re-branding, I think it's still net-positive.
  - rramadass 2y ago
    Well said! As you point out; "making people think in more novel ways" (or making them think at all given the amount of parroting in the education system) is the usp of Taleb's books. People need to stop focusing on the attributes of the author (arrogant etc.) but instead need to learn to focus on the content of their writings.
mo_42 2y ago
I'd like to point out that Norbert Wiener was the first to discover the concept of antifragility (under a different term though). It's also worth checking out more of his works as he initiated the field of cybernetics.
- agumonkey 2y ago
  Where did he described this? (Honest question)
  - mo_42 2y ago
    To be correct, he maybe didn't coin the term. At least I didn't find it on quick search in his book Cybernetics. Seems like the term ultrastable actually comes from Ashby [1]. [1] https://users.sussex.ac.uk/~ezequiel/AS/lectures/AdaptiveSystems3.pdf https://users.sussex.ac.uk/~ezequiel/AS/lectures/AdaptiveSys...
    - agumonkey 2y ago
      Yeah, I'm more curious about his point of view than terminology. Thanks
  - r_hanz 2y ago
    I believe Nassim Taleb coined the term.
fasteo 2y ago
Random note: Antifragility is called hormesis in living organisms, a concept that existed since the 1950s when the wrong dose - too low - of herbicides made the plants stronger.
- Beretta_Vexee 2y ago
  Mechanical engineering uses the term system resilience. I must admit I don't like antifragility, which sounds like a novlang and is poorly defined.
  - dmichulke 2y ago
    Resilience is not the same, a system is anti-fragile if it gets stronger under stress. Resilience is just "not getting weaker".
    - K0balt 2y ago
      Pedantically-not pedantic, but isn’t resilience the ability to quickly recover?
      - m0llusk 2y ago
        The first graphic in the paper has a good illustration of the differences between fragility, robustness, resilience, adaptiveness, and antifragility: https://www.nature.com/articles/s44260-024-00014-y/figures/1 https://www.nature.com/articles/s44260-024-00014-y/figures/1
      - fuzzfactor 2y ago
        When enduring successive hurricanes the most worthwhile effort might be to improve antifragility on more of a continuous basis than even recovery itself. You can definitely benefit by increasing your antifragility in the face of a second hurricane long before recovery from the first hurricane has been very well completed at all, no matter how focused the recovery effort. I never thought people needed a new term to be coined for this, but it doesn't sound so bad when you think about it. It's just one of the things you inherently know your whole life.
cfcf14 2y ago
After reading the paper, I'm really unsure what the novel contribution is. It feels like they're attempting to rebrand well-understood concepts within various fields (control systems theory, etc). The provided mathematical definition of antifragility is somewhat unconvincing too: it's not that it's wrong, per say, but in the effort to find something sufficiently broad to apply to many different fields of applied dynamical theory they've had to adopt a definition which is a bit unintuitive, and overly general.
lokimedes 2y ago
How odd that the word "Antifragility" has risen so fast. What's wrong with robust? Is it due to contemporary memes of personal fragility?
- brians 2y ago
  It’s trying to highlight a different idea. A robust system can endure damage. An antifragile system grows stronger from damage.
- buggythebug 2y ago
  Title of a best seller by a respectable author, Nassim Taleb
- Eddy_Viscosity2 2y ago
  I, along with several other commenters I see, also dislike "antifragility" as a word. However, Taleb in his book does discusses at length why a new word was needed and how its definition differs from existing words like 'robust', 'resilient', etc. which have a different meaning. He even talks even the process of choosing "antifragility" which in a nutshell was just the best he could come up with. Seeing other comments, I think 'hormesis' might be a better word to start using instead of "antifragility" which, again, is a just an awkward word that feels kind of fake when you use it.

173 points by RafelMri 2y ago

I've found that intentionally causing abrupt (but reasonable) changes to hyperparameters in evolutionary spiking neural network simulations (I.e between each generation) results in far more robust simulations that will meet fitness criteria with less likelihood of getting stuck somewhere. The tradeoff being that simulating will take longer, but this may be worth things like reducing the chances of your resource requirements going asymptotic half way through. My current method of perturbation is to cap the total # of activations per candidate globally. When at least 3 improvements in the best global fitness score have been achieved, I cut the global activation limit in half. Obviously, this is fairly catastrophic for the current population. Each generation can increase the limit by 1% if no improvements made. This provides aggressive selection pressure for more efficient networks and forces a different kind of partial restart of the training process (I.e. do what you just did with 50% resources now). Very often this does result in optimization of the network. It also seems to make the networks rebound faster after restarts. At first, a restart can take 10-20 generations to recover. By generation 1000, this is closer to 1-2.

_yb2s 2y ago
Very cool, but I'm surprised you're sharing this on here and not in a job interview with a deep learning startup and/or an arXiv paper.
- bob1029 2y ago
  > a job interview with a deep learning startup I am not aware of any DL startups that are interested in techniques which are incompatible with GPUs and back propagation. I think this is a solo journey through the dark forest for now.
  - inciampati 2y ago
    Eloquent, realistic, and sad. I'm living my own version of this (in biotech). The journey is the wonder.
  - phyalow 2y ago
    Correct me if im wrong, this is just a question originating from my own curiosity, but could you not apply the same or similar techniques to classical NN architectures? I dont see exactly how the differentiability of the architecture matters? I.e. in between epochs randomly perturb weight values (similar effect I understand to introducing scheduled noise or non-isomorphic transforms on the features), implement a scheduled causal drop in/out scheme etc? I know of people that train models these ways as a form of regularisation and they claim better fits too. Enjoy your adventures in the Dark Forest.
    - bob1029 2y ago
      > I dont see exactly how the differentiability of the architecture matters? I don't know that classical architectures are infeasible to perturb as noted in other comments. I just think we have a really interesting knob to turn in this specific case that doesn't seem to exist in other architectures. How would you go about constraining the resource consumption (by up to half!) of a classical ANN while preserving realistic hopes that it will recover between training epochs? And, we are talking real resource reduction here not some academic thing. You can see the difference in task manager. That's actually the primary tool I've been using to determine how to tune the parameters of this scheme.
      - SuchAnonMuchWow 2y ago
        There has been some work to dynamically reduce the compute required by a network. See for example: https://arxiv.org/abs/2404.02258 https://arxiv.org/abs/2404.02258 They have a fixed compute budget which is lower than what the LLM need, and dynamically decide to allocate this compute budget to different part of the network. So its not exactly what you propose since here the compute budget is fixed (that's the point of the paper: to make the network learn how to allocate the resources by itself) but its dynamic for each part of the network, so it shows that its possible.
  - sva_ 2y ago
    > I think this is a solo journey through the dark forest for now. In academia there's quite a bit of stuff on neuromorphic computing going on.
- sethammons 2y ago
  Your surprise breaks my brain. Why would someone "keep it to themselves" outside some white paper they won't have time to promote or for a job interview that they might not need? Is your thought that it is some novel approach that nobody is doing and has obvious market implications? Even then, the implications matter only if you are in a position to take advantage. The Internet used to be a place people shared cool shit because it was cool, not for some monetization scheme or personal gain beyond the joy of sharing. Content creators (as a job) and hustle culture are a blight.
  - baxtr 2y ago
    They are sharing it here?! It’s not like OP is keeping it to themselves
  - _yb2s 2y ago
    As I'm sure you're aware deep learning work better is the most hot and highly funded thing right now... there is nothing wrong with making sure you get the credit for a new technique, and it in no way precludes widespread free sharing of the idea. I am an academic researcher (not in this field) and everything I do is 100% free and public, but I still either patent or publish it, because getting the credit is what allows me to keep doing what I love. I actually hate that part, and would rather just do the work, but I already push that line too far sometimes, and risk losing my funding aka "publish or perish."
- authorfly 2y ago
  Respect to the OP as spiking networks are a different domain, but the field of Graceful Degradation and Perturbations itself is almost older than the field of Neural Networks itself. Generally, the bitter lesson occurs - these techniques show some kind of useful robustness at parameter M, but fail to show as much of a useful robustness relatively as parameter M*100. For example, before attention, using Drop out (in essence a form of forcing robustness) was one of the most common techniques. The quality of robustness or its resulting sub parts (e.g. the tradeoff between precision and recall) is incorporated as a measure in some of the fields basic scores, such as harmonic F1 score. Still it, like alternate activation functions, data beyond matrices and vectors and non-linear networks, remain very unstudied.
JackeJR 2y ago
I am quite curious about your approach, Do you have an example code somewhere?
- bob1029 2y ago
  I'd start here: https://arxiv.org/abs/2406.12552 https://arxiv.org/abs/2406.12552 I will say that less is more and "The Bitter Lesson" applies here. Chasing biologically-inspired rabbits, such as STDP and LIF (see paper above/wikipedia), does seem to be a waste of time, especially when we have this thing entirely outside of biology that can arbitrarily replicate, serialize, mutate and simulate billions of instances of the same parent candidate in minutes-hours. Leaky charge carriers and inability to persist learned weights between candidate instantiations are limitations, not features to emulate. Imagine if you could be reincarnated with all of the exact knowledge you have today. Then clone that 1000 times and apply subtle mutations to it. Then, put these clones in an FFA arena and see who wins based upon a very well crafted fitness function. Then, do all of that over and over thousands of times per hour.
  - jcims 2y ago
    >Chasing biologically-inspired rabbits, such as STDP and LIF (see paper above/wikipedia), does seem to be a waste of time Unless physiological compatibility with a biological brain is considered an interesting endpoint? If you think about Neuralink for example, wouldn’t it be interesting if our brains could directly engage the model? Not just a translation layer but perceive it directly and natively though some kind of synaptic modem that converts analog exchanges of neurotransmitters to the synthetic network in the digital domain.
    - bob1029 2y ago
      > Unless physiological compatibility with a biological brain is considered an interesting endpoint? Excellent point. I am focused entirely on synthetic simulations without biological integration (for now). I think that could be an interesting next step - determining a way to modify a SNN that was trained in a synthetic time domain such that it can process real time signals. Training these things online is not really an option. You have to run a large # of offline simulations before you can find the right networks. Learning rules like STDP could theoretically address the online learning problem, but I couldn't find any traction down that path yet.
taylorius 2y ago
Firstly, I'm coming at this from the pint of view of standard feed forward NNs, rather than spiking neural networks, so apologies in advance if I am way off-base. By global activation limit, I assume you mean the total sum of neuron activation during one forward pass? So halving it would strongly encourage sparsity? How do you enforce the reduced limit? Just put it in the loss function? Or does the spiking nature allow you to "sparsify" the spiking by raising a threshold? Very very interesting stuff. I'm working with differential evolution, and incorporate correlation measures between candidates to (try to) allow for multiple divergent solutions, and it also feels fairly off the beaten track, compared to mainstream AI. But who cares, let's enjoy the dark forest! :-)
- bob1029 2y ago
  > By global activation limit, I assume you mean the total sum of neuron activation during one forward pass? Yes. For each candidate network I keep a counter of the total # of neuron activations. This figure is constrained by a global limit (which varies based on prior performance). We simply break out of the simulation loop once the allotted resources are exhauted. This means that more efficient networks are more likely to score higher.
seanhunter 2y ago
> I've found that intentionally causing abrupt (but reasonable) changes to hyperparameters in evolutionary spiking neural network simulations (I.e between each generation) results in far more robust simulations that will meet fitness criteria with less likelihood of getting stuck somewhere. If I understand you correctly this is similar to the philosophy behind simulated annealing isn't it? The idea being that adding some peturbation can be enough to jog the evolution out of local minima and you probably want to reduce the amount of those jogs over time so it ends up converging. https://en.wikipedia.org/wiki/Simulated_annealing https://en.wikipedia.org/wiki/Simulated_annealing
- neom 2y ago
  Not op but what they described is indeed sim annealing coupled with relaxing feedback loops. This approach would in theory result in prevention of stagnation, forced adaptation and over time, creating some form of resiliency. Pretty cool stuff!
- bob1029 2y ago
  I was actually thinking about simulated annealing as I got into this. I've found that reducing this specific mode of perturbation over time has not been necessary. The simulation will oscillate its way right into a perfect fitness score assuming everything else is good. Convergence seems to rely much more heavily on the quality of the fitness function and population management (i.e. size, selection criteria, mutation strategies). I apply random mutations to delays and weights, but using a spectrum of mutation each time (I.e. some clones are heavily mutated while others are barely touched). This spectra of mutation is constant all the way through the simulation.
lo0dot0 2y ago
Some of you may die... But that's a sacrifice i'm willing to make.

adiM 2y ago

Is this just risk seeking behavior (in contrast to risk sensitive)

marci 2y ago
Now I wonder what a machine learning model with risk seeking behaviour would look like.
- mike_ivanov 2y ago
  https://news.ycombinator.com/item?id=41231644 https://news.ycombinator.com/item?id=41231644
  - marci 2y ago
    I had read this comment, but didn't read it under this lens. Thanks for (re)pointing me to it.
- nottorp 2y ago
  From the little I know about ML in general, you always want to introduce randomness so you don't get stuck in a local minimum/maximum. So they're all risk seeking?

delichon 2y ago

Systems or organisms can be defined as antifragile if they derive benefit from systemic variability, volatility, randomness, or disorder. If that's a riddle, death fits. Those things are characteristic of the pestilence, famine, war, etc. that feed death. The decay surrounding death is complex and dynamical. And death is more sustainable than any living system.

_yb2s 2y ago
but death isn't a system or organism...
- fsckboy 2y ago
  >but death isn't a system death is entropy. do you know of a system without entropy? >or organism... boy are you going to be surprised when you answer that knock on the door, eh?
  - BriggyDwiggs42 2y ago
    Isn’t it definitely not an organism? If death is the inevitable process from order to disorder, isn’t it deeply a-systemic at its core? If it were a system, it would also eventually have to end.
    - hobs 2y ago
      Arguably it does, the heat death of the universe option would spell the end of entropy, and the end of death (unless you believe in a conformal cyclic universe)
      - BriggyDwiggs42 2y ago
        Actually yeah fair enough; I’ll grant that death may be a system.
        _yb2s 2y ago

AgentOrange1234 2y ago

I wish folks would write more plainly. “Antifragility characterizes the benefit of a dynamical system derived from the variability in environmental perturbations.” Geish.

_yb2s 2y ago
Well it is in a journal called "complexity." But I agree, that sentence is awful, I'm surprised they decided to lead with it, and nobody proof reading the paper objected.
- defrost 2y ago
  The leading three authors are from Germany, Mexico, and Switzerland resptfully. The sentence they want should capture the notion of being robust when poked with a stick. A pencil on it's tip is a fragile system, one burp and it falls to the table, far from the intial state. Marbles in fruit bowls are anti-fragile, given a good shake (up to a threshold) and they remain in the bowl and return to the low centre. Reading further, they want more; that repeated perturbations should deliver benefit, that systems in an warped egg carton configuration can be annealed to reveal an optimal point by vigorous shaking slowly reduced in degree. Having now read the whole paper it doesn't seem novel wrt to the state of dynamic system discussions in the mid 1980s other than the insertion of "antifragility" as nomenclature into the mix.
  - wenc 2y ago
    > A pencil on it's tip is a fragile system, one burp and it falls to the table, far from the intial state. Marbles in fruit bowls are anti-fragile, given a good shake (up to a threshold) and they remain in the bowl and return to the low centre. I believe what you're describing is a stable vs unstable system, not fragile vs antifragile. You can perturb a bowl with a marble, and the marble will still end up in the middle because it will return to stable equilibrium point (an "attractor"). Yours is an illustration of a stable system. Whereas a marble placed in an upside down bowl (with no ridges, just a half sphere), when perturbed, will fall off. This is an unstable system. These are classic examples used in (Lyapunov) stability theory. Fragility and antifragility aren't about stability (returning to equilibria), but gains or losses after perturbation, which is related to convexity/concavity. When you perturb an anti-fragile (or convex) system, it doesn't return to equilibrium but in fact improves. Conversely, when you perturb a fragile system, it degrades. The analysis is usually done with Jensen's inequality rather than Lyapunov. EDIT: not sure why the downvotes. I'm pointing out a fact. The examples do not demonstrate antifragility, but stability, which is not the same concept.

_yb2s 2y ago

I love how Taleb managed to translate an ideal from stoic philosophy into a precisely defined mathematical concept

Animats 2y ago
Some of this reads like rediscovering control theory, which is all about stability and robustness within defined limits. It's more popularization than innovation. The person who first translated this idea into math was James Clerk Maxwell, in his paper, "On Governors", in 1868.[1] Maxwell was the first to get a mathematical handle on stability of feedback systems. He wrote: "If, by altering the adjustments of the machine, its governing power is continually increased, there is generally a lin1it at which the disturbance, instead of subsiding more rapidly, becomes an oscillating and jerking motion, increasing in violence till it reaches the limit of action of the governor. This takes place when the possible part of one of the impossible roots becomes positive. ... I have pointed out that, under certain conditions, the sudden disturbances of the machine do not act through the differential system on the governor, or vice versa. When these conditions are fulfilled, the equations of motion are not only simple, but the motion itself is not liable to disturbances depending on the mutual action of the machine and the governor." That, right there, is the beginning of the mathematics of antifragility. Today, we call this turning up the P term of a PID controller too high and getting outside the stable region. Maxwell was the first to formalize the role of delay and damping (he calls it "viscosity") in this. There's "robust control", a useful theory of "antifragility" in industrial use for decades. Here's an excellent video on how this is done.[2] That's worth a watch. Note the emphasis on how the combination of delay and noise can cause instability in a system resistant to both delay and noise separately. This can be dealt with by moving the areas of instability around in the frequency domain. (Modern control theory has become insanely complicated over the last few years. Until recent years, control theory was mostly linear control theory plus some self tuning. Now, control theorists are trying to figure out how to get the benefits of machine learning while still having predictable error bounds. This means getting into the innards of what neural nets are really doing. I still get IEEE Trans. on Control Systems Technology, but don't understand it any more. So I'm not sure how that's coming along.) [1] https://archive.org/details/jstor-112510/mode/2up?view=theater https://archive.org/details/jstor-112510/mode/2up?view=theat... [2] https://www.youtube.com/watch?v=A7wHSr6GRnc https://www.youtube.com/watch?v=A7wHSr6GRnc

krick 2y ago

There are now so many bullshit-terms derived from Taleb's bullshit books. Like how people still keep mentioning "black swans" as if it actually means something other (or something more) than "unexpected event". And for some unfathomable reason it keeps traction. Similarly how Mandelbrot "redefined" (i.e. distorted) the meaning of "Lindy effect", and it stuck (however, I didn't notice if it became popular to call a millenia-old banality by than name after Mandelbrot, or Taleb again). It probably should serve as an another example of that if you are arrogant enough, people will follow you just because of that. However, I guess "antifragility" isn't the worst of these.

aaron695 2y ago
[dead]
nottorp 2y ago
When I saw the title I thought it came out of one of those online bullshit generators...
padolsey 2y ago
I don't much mind the populurization of these terms or making people think in more novel ways. Even if it's seen as BS by those more informed, I'm glad such books are written and communicators like Taleb exist. Without him, I wouldn't have discovered a bunch of tangential things. I will admit, it gives my brain a satisfying itch too, as I realise that academia is often just the refined encoding of pretty mundane everyday truths, so when someone is able to come in and re-extract that and share it widely, even with a bit of gentle-re-branding, I think it's still net-positive.
- rramadass 2y ago
  Well said! As you point out; "making people think in more novel ways" (or making them think at all given the amount of parroting in the education system) is the usp of Taleb's books. People need to stop focusing on the attributes of the author (arrogant etc.) but instead need to learn to focus on the content of their writings.

mo_42 2y ago

I'd like to point out that Norbert Wiener was the first to discover the concept of antifragility (under a different term though). It's also worth checking out more of his works as he initiated the field of cybernetics.

agumonkey 2y ago
Where did he described this? (Honest question)
- mo_42 2y ago
  To be correct, he maybe didn't coin the term. At least I didn't find it on quick search in his book Cybernetics. Seems like the term ultrastable actually comes from Ashby [1]. [1] https://users.sussex.ac.uk/~ezequiel/AS/lectures/AdaptiveSystems3.pdf https://users.sussex.ac.uk/~ezequiel/AS/lectures/AdaptiveSys...
  - agumonkey 2y ago
    Yeah, I'm more curious about his point of view than terminology. Thanks
- r_hanz 2y ago
  I believe Nassim Taleb coined the term.

fasteo 2y ago

Random note: Antifragility is called hormesis in living organisms, a concept that existed since the 1950s when the wrong dose - too low - of herbicides made the plants stronger.

Beretta_Vexee 2y ago
Mechanical engineering uses the term system resilience. I must admit I don't like antifragility, which sounds like a novlang and is poorly defined.
- dmichulke 2y ago
  Resilience is not the same, a system is anti-fragile if it gets stronger under stress. Resilience is just "not getting weaker".
  - K0balt 2y ago
    Pedantically-not pedantic, but isn’t resilience the ability to quickly recover?
    - m0llusk 2y ago
      The first graphic in the paper has a good illustration of the differences between fragility, robustness, resilience, adaptiveness, and antifragility: https://www.nature.com/articles/s44260-024-00014-y/figures/1 https://www.nature.com/articles/s44260-024-00014-y/figures/1
    - fuzzfactor 2y ago
      When enduring successive hurricanes the most worthwhile effort might be to improve antifragility on more of a continuous basis than even recovery itself. You can definitely benefit by increasing your antifragility in the face of a second hurricane long before recovery from the first hurricane has been very well completed at all, no matter how focused the recovery effort. I never thought people needed a new term to be coined for this, but it doesn't sound so bad when you think about it. It's just one of the things you inherently know your whole life.

cfcf14 2y ago

After reading the paper, I'm really unsure what the novel contribution is. It feels like they're attempting to rebrand well-understood concepts within various fields (control systems theory, etc). The provided mathematical definition of antifragility is somewhat unconvincing too: it's not that it's wrong, per say, but in the effort to find something sufficiently broad to apply to many different fields of applied dynamical theory they've had to adopt a definition which is a bit unintuitive, and overly general.

lokimedes 2y ago

How odd that the word "Antifragility" has risen so fast. What's wrong with robust? Is it due to contemporary memes of personal fragility?

brians 2y ago
It’s trying to highlight a different idea. A robust system can endure damage. An antifragile system grows stronger from damage.
buggythebug 2y ago
Title of a best seller by a respectable author, Nassim Taleb
Eddy_Viscosity2 2y ago
I, along with several other commenters I see, also dislike "antifragility" as a word. However, Taleb in his book does discusses at length why a new word was needed and how its definition differs from existing words like 'robust', 'resilient', etc. which have a different meaning. He even talks even the process of choosing "antifragility" which in a nutshell was just the best he could come up with. Seeing other comments, I think 'hormesis' might be a better word to start using instead of "antifragility" which, again, is a just an awkward word that feels kind of fake when you use it.