Showing posts with label probability. Show all posts
Showing posts with label probability. Show all posts

The Coming Crisis: Fingers of Instability

Mike's Notes

An excellent description of what a critical state is, using a real-world example: the global financial system. Critical state also applies to other phenomena, including earthquakes.

I have read the two excellent books by Buchanan and Taleb in the references. I must also read Sornette's book.

I think everything in the universe has its time in the sun, with a birth, existence, and death, often followed by a transformation into its oppositeThe laws of science apply to everything, including social systems like capitalism, trees, planets, schools of music, cars, etc.

Resources

References

  • Why Catastrophes Happen, by Mark Buchanan.
  • Antifragility, by Nassim Taleb.
  • Why Stock Markets Crash, by Didier Sornette.

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Thoughts from the Frontline
  • Home > Handbook > 

Last Updated

02/03/2026

The Coming Crisis: Fingers of Instability

By: John Maudlin
Thoughts from the Frontline: 28/02/2026

John Maudlin is Co-Founder, Mauldin Economics.

This letter is a little different. I am indeed working on my book about what I believe is a coming crisis by reviewing five different cycle theories. They all arrive at a similar scenario from  different points of view, but they all suggest a crisis occurring sometime around the end of this decade or perhaps shortly thereafter. And all for different reasons. One background element ties them together, which is the subject of today’s letter.

This is essentially a shortened first chapter. To long time readers, that background connection is our old friend: sandpiles and fingers of instability. but with a lot of edits and additions. Jumping in…

Ubiquity, Complexity Theory, and Sandpiles

With five different views about the coming crisis, which one is right? Do they conflict or reinforce each other? The correct answer is they’re all connected, but not in obvious ways. And in the end, it makes no difference which one is “more” right. The results will be the same. Understanding this below-the-radar connection is key to making sure you, your family, community and country all get through this to what will be the inevitable positive conclusion, even if it is a very bumpy ride.

We are going to start our exploration with excerpts from an important book by Mark Buchanan, called Why Catastrophes Happen. I HIGHLY recommend it to those of you who, like me, are trying to understand the complexity of the markets, economy and politics/society. The book is about chaos theory, complexity theory and critical states. It is written in layman’s terms. There are no equations, just easy-to-grasp, well-written stories and analogies. But it gives us an essential framework to understand the coming storms.

As kids, we all had the fun of going to the beach and playing in the sand. Remember taking your plastic buckets and making sand piles? Slowly pouring the sand into an ever-bigger pile, until one side of the pile started an avalanche?

Imagine, Buchanan says, dropping one grain of sand after another onto a table. A pile soon develops. Eventually, just one grain starts an avalanche. Usually it’s a small one, but sometimes it builds on itself and seems like a side of the pile collapses. Why?

Well, in 1987 three physicists named Per Bak, Chao Tang, and Kurt Weisenfeld began to play the sandpile game in their lab at Brookhaven National Laboratory in New York. Now, piling one grain of sand at a time is a slow process, so they wrote a computer program to do it. Not as much fun, but a whole lot faster. Not that they really cared about sandpiles. They were interested in what are called nonequilibrium systems.

They learned some interesting things. What is the typical size of an avalanche? After a huge number of tests with millions of grains of sand, they found there is no typical size. "Some involved a single grain; others, ten, a hundred or a thousand. Still others were pile-wide cataclysms involving millions that brought nearly the whole mountain down. At any time, literally anything, it seemed, might be just about to occur." The piles were chaotic in their unpredictability.

Now, let’s read this next paragraph from Buchanan slowly. It is important, as it creates a mental image that may help us understand the organization of financial markets, the world economy and society (emphasis mine).

"To find out why (such unpredictability) should show up in their sandpile game, Bak and colleagues next played a trick with their computer. Imagine peering down on the pile from above, and coloring it in according to its steepness. Where it is relatively flat and stable, color it green; where steep and, in avalanche terms, ‘ready to go,’ color it red. What do you see? They found that at the outset the pile looked mostly green, but that, as the pile grew, the green became infiltrated with ever more red. With more grains, the scattering of red danger spots grew until a dense skeleton of instability ran through the pile. Here then was a clue to its peculiar

The Critical State

Something only a math nerd could love? Scientists refer to this as a “critical state.” The term can mean the point at which water goes to ice or steam, or the moment that critical mass induces a nuclear reaction, etc. It is the point at which something triggers a change in the basic nature or character of the object or group. Thus (and very casually for all you physicists), we refer to something being in a critical state (or use the term critical mass) when there is the opportunity for significant change.

"But to physicists, [the critical state] has always been seen as a kind of theoretical freak sideshow, a devilishly unstable and unusual condition that arises only under the most exceptional circumstances [in highly controlled experiments]… In the sandpile game, however, a critical state seemed to arise naturally through the mindless sprinkling of grains."

Thus, they asked themselves, could this phenomenon show up elsewhere? In the earth’s crust, triggering earthquakes, or as wholesale changes in an ecosystem – or as a stock market crash?

"Could the special organization of the critical state explain why the world at large seems so susceptible to unpredictable upheavals?" Could it help us understand not just earthquakes, but why cartoons in a third-rate paper in Denmark could cause world-wide riots?

Buchanan concludes in his opening chapter:

"There are many subtleties and twists in the story … but the basic message, roughly speaking, is simple: The peculiar and exceptionally unstable organization of the critical state does indeed seem to be ubiquitous in our world. Researchers in the past few years have found its mathematical fingerprints in the workings of all the upheavals I’ve mentioned so far [earthquakes, eco-disasters, market crashes], as well as in the spreading of epidemics, the flaring of traffic jams, the patterns by which instructions trickle down from managers to workers in the office, and in many other things. At the heart of our story, then, lies the discovery that networks of things of all kinds – atoms, molecules, species, people, and even ideas – have a marked tendency to organize themselves along similar lines. On the basis of this insight, scientists are finally beginning to fathom what lies behind tumultuous events of all sorts, and to see patterns at work where they have never seen them before."

Going back to the sandpile game, you find that as you double the number of grains of sand involved in an avalanche, the probability of an avalanche becomes 2.14 times more likely. We find something similar in earthquakes. In terms of energy, the data indicate that earthquakes become four times less likely each time you double the energy they release. Mathematicians refer to this as a "power law," a special mathematical pattern that stands out in contrast to the overall complexity of the earthquake process.

Fingers of Instability

So, what happens in our game?

"…after the pile evolves into a critical state, many grains rest just on the verge of tumbling, and these grains link up into ‘fingers of instability’ of all possible lengths. While many are short, others slice through the pile from one end to the other. The chain reaction triggered by a single grain might lead to an avalanche of any size whatsoever, depending on whether that grain fell on a short, intermediate or long finger of instability."

Now, we come to a critical point in our discussion of the critical state. Again, read this with not just markets but our entire society in mind:

"In this simplified setting of the sandpile, the power law also points to something else: the surprising conclusion that even the greatest of events have no special or exceptional causes. After all, every avalanche, large or small, starts out the same way, when a single grain falls and makes the pile just slightly too steep at one point. What makes one avalanche much larger than another has nothing to do with its original cause, and nothing to do with some special situation in the pile just before it starts. Rather, it has to do with the perpetually unstable organization of the critical state, which makes it always possible for the next grain to trigger an avalanche of any size."

This concept applies to not just financial markets, but to how we organize our political systems, generational differences, geopolitics and war, the over-production of elites and even how information is interpreted. They ALL connect. The Great Recession was a financial crisis. COVID-19 was a health crisis with a financial crisis and added political crises which further divided a fractious world.

We all see pressures building up in many different aspects of society. They each create their own fingers of instability. But in the sandpile of life, they are connected. 

Now, let’s couple this idea with a few other concepts. First, Hyman Minsky (who should have been a Nobel laureate) points out that stability leads to instability. The more comfortable we get with a given condition or trend, the longer it will persist and then when the trend fails, the more dramatic the correction.

The problem with long term macroeconomic stability is that it tends to produce unstable financial arrangements. Just as long term geopolitical or social stability will eventually produce a critical state. If we believe that tomorrow and next year will be the same as last week and last year, we are more willing to add debt or postpone savings in favor of current consumption. Or ignore any of a number of societal crises. Thus, says Minsky, the longer the period of stability, the higher the potential risk for even greater instability when market participants or a country’s citizens must change their behavior.

Relating this to our sandpile, the longer a critical state builds up in an economy, or in other words, the more "fingers of instability" are allowed to develop connections to other fingers of instability, the greater the potential for a serious "avalanche."

Therefore (and ironically), the longer a crisis takes to come about, the bigger the repercussions. One of the conclusions at the end of the book will be that we simply don’t know when the avalanche will be triggered. The US is such a large and wealthy country, and many of the rest of the shirts in the global laundry are just as (or even more) dirty, that global money might come to the US as a safe haven, thus prolonging our “stability” as the sandpile grows to an ever more critical state.

We Are Managing Uncertainty

Or, maybe, a series of smaller shocks lessens the long reach of the fingers of instability, giving a paradoxical rise to even more apparent stability. This is the thrust of Nassim Taleb’s book, Antifragility.

“People often think that the opposite of fragility is durability. If something is fragile, that means it’s easily broken. Therefore, if something isn’t easily broken, logically that should mean it’s the opposite of fragile. However, there’s another step beyond. Since there isn’t an established English word for such a thing, [Nassim] calls it antifragility—not just the lack of fragility, but its true opposite.

“We live in an unpredictable world. The models and theories we use to try to predict the future invariably fall apart as unforeseen events prove them wrong and, in turn, destroy the plans we made based on those models. Clearly, systems based on such flawed models are bound to be fragile—easily broken.

“The solution to this problem is antifragility. Instead of a never-ending search for more accurate models and better predictions, all we need to do is make sure that we’re in a position to benefit from uncertainty and volatility instead of being harmed by it.

“This is hardly a new concept; nature exhibits antifragility in almost everything she creates. An organism can strengthen itself through minor damage in the form of exercise. In a similar sense, a species can strengthen itself through minor damage in the form of natural selection, which leads to evolution.

“However, unlike nature, humans try to control the world through models and rules. We think we can perfectly predict the future and avoid any shocks that would cause our fragile systems to fall apart. We think we can outsmart millions of years of evolution and antifragility, and we’re almost invariably wrong.

“Instead of trying to predict the future, we should assume that there will be major events we can’t see coming—because, sooner or later, there will be. If we’re prepared for them, using the methods and practices explained in this book, we can make sure that such events work to our advantage instead of hurting us. By avoiding fragility and embracing antifragility wherever possible, we can set ourselves up to thrive in an uncertain world.

Another way to think about it is the way Didier Sornette, a French geophysicist, has described financial crashes in his wonderful book, Why Stock Markets Crash (the math, though, was far beyond me!). He wrote:

"[T]he specific manner by which prices collapsed is not the most important problem: a crash occurs because the market has entered an unstable phase and any small disturbance or process may have triggered the instability. Think of a ruler held up vertically on your  the instantaneous cause of the collapse is secondary."

When things are unstable, it isn’t the last grain of sand that causes the pile to collapse or the slight breeze that causes the ruler on your fingertip to fall. Those are the "proximate" causes. They’re the closest reasons at hand for the collapse. The real reason, though, is the "remote" cause, the farthest reason. The farthest reason is the underlying instability of the system itself.

This is one reason we get "fat tails" in financial markets. In theory, returns on investment should look like a smooth bell curve, with the ends tapering off into nothing. According to the theoretical distribution, events that deviate from the mean by five or more standard deviations ("5-sigma events") are extremely rare, with 10 or more sigma being practically impossible – at least in theory.

However, under certain circumstances, such events are more common than expected; 15-sigma or even rarer events have happened in the world of investing. Examples include Long Term Capital in the late 1990s and any of a dozen bubbles in history. Because the real-world commonality of high-sigma events is much greater than in theory, the distribution is "fatter" at the extremes ("tails") than one would expect.

This holds true in geopolitics, too. The unthinkable sometimes happens. Before World War I began, no one thought it would come to war. Peace had been the rule for 40 years. Surely, mankind had evolved. Until…

Thus, the build-up of critical states, those fingers of instability, is perpetuated even as, and precisely because, we hedge risks. We try to "stabilize" the risks we see, shoring them up with derivatives, emergency plans, insurance, treaties, alliances, political change and all manner of risk-control procedures. And by doing so, the economic and social systems can absorb body blows that would have been severe only a few decades ago. We distribute the risks, and their effects, throughout the system.

Yet as we reduce the known risks, we sow the seeds for the next 10-sigma event. It is the improbable, unseen risks that will create the next real crisis. It is not that the fingers of instability have been removed from the equation, it is that they lurk in different places, not yet visible.

A Stable Disequilibrium

We end up in a critical state that Paul McCulley calls "stable disequilibrium." It has "players" all over the world, tied inextricably together in a vast dance through investment, debt, derivatives, trade, globalization, international business and finance. Each player works hard to maximize their own personal outcome and reduce their exposure to "fingers of instability."

The longer we go on, asserts Minsky, the more likely and violent any "avalanche" is. The more the fingers of instability can build, the more that state of stable disequilibrium can go critical on us.

It's all connected. We are building an unstable sandpile and it will come crashing down at some point. Then we will have to dig our way out.

The good news is we have seen this movie before. And after the crisis, a new period of stability and growth follows, for at least another 50-80 years. In my upcoming book we will look for ways to get through to that happier future.

Scottsdale, Houston, Los Angeles, West Palm Beach, Boston and New York

Next week I fly to Houston where I am on an economic advisory board for the Rice University economics department. Then I will be in LA meeting with the Inner Circle, exploring several companies that are literally changing the technology landscape of defense and energy. We will be opening clinics in West Palm Beach and the DC area, hopefully in early April. Construction has begun. Then NYC and Boston.

I finish this from Scottsdale where Dr. Roizen and I are attending the 2026 Functional Longevity Summit, along with 3-400 doctors. The organizers have asked us to talk about Therapeutic Plasma Exchange. For those interested in staying healthy for longer, Mike and many experts now believe the first part of your journey should begin with therapeutic plasma exchange. Seriously. You can learn more at Lifespan-Edge.com (note the dash). If you haven’t, you really need to read our main research report. The research and other information can make a real difference in your life. You can set up a discovery call to talk with our doctors about the procedure and see if it is right for you. As well as look at a lot more research.

And with that, I will hit the send button. Have a great week.

Your thinking how to make my body antifragile analyst,

Understanding the Hidden Markov Model

Mike's Notes

A Markov Chain is a collection of states with transition probabilities between them.

Markov is used extensively in Pipi 9. Here is some excellent background explanations from Vivek Vinushanth Christopher on Built In.

Vivek is a fantastic teacher with an excellent YouTube channel on Data Science.

While the core code of Pipi will be closed source, the weights, algorithms, possible states, ontologies used, and related materials will be made freely available. The front-end workspaces will be shared on GitHub, GitLab, and other platforms.

The planned user documentation will include references to popular books, articles, and easy-to-understand videos from people like Vivek.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library >
  • Home > Handbook > 

Last Updated

24/01/2026

Understanding the Hidden Markov Model

By: Vivek Vinushanth Christopher, Updated By: Brennan Whitfield
Built In: 14/08/2024

Vivek Vinushanth Christopher is a senior software engineer for WSO2 where he has worked since 2019. He has also served as a teaching assistant at University of Moratuwa. Christopher holds a bachelor’s degree in computer science and engineering. 

Hidden Markov models are probabilistic frameworks where the observed data are modeled as a series of outputs generated by one of several (hidden) internal states. Both Markov and hidden Markov models are engineered to handle data that can be represented as a sequence of observations over time.

Hidden Markov Model Definition

A hidden Markov model is a probabilistic framework used to predict the results of an event based on a series of observations with one or several hidden internal states.

While a Markov model or Markov chain concerns stochastic (random) process states that are visible to the observer, a hidden Markov model pertains to stochastic processes where states can be hidden or not directly visible to the observer.

What Is a Hidden Markov Model (HMM)?

A hidden Markov model (HMM) is utilized when we can’t observe the states of a stochastic process themselves, but only the result of some probability function (observation) of the states. HMM is a statistical Markov model in which the system being modeled is assumed to be a Markov process with unobserved (hidden) states.

Hidden Markov models can be used to identify underlying patterns or structures in sequential data. This makes it applicable for research and tasks in machine learning (including natural language processing and speech recognition), bioinformatics and gene analysis as well as time-series forecasting. 

For example, in speech recognition tasks, a hidden Markov model algorithm may be implemented to measure the probability of a certain word or lack of words occurring in a given audio recording. In this case, the occurrence of specific words or silence in the recording can represent states, and volume of speech throughout the recording can represent observations. By knowing observations (volume), this information can be used by the algorithm to determine the likelihood of hidden states — or words and lack of words in this example — and predict the most probable word being spoken.

Mathematically, here is how Markov models and hidden Markov models differ:

  • Markov model: Series of (hidden) states z={z_1,z_2………….} drawn from state alphabet S ={s_1,s_2,…….𝑠_|𝑆|} where z_i belongs to S.
  • Hidden Markov model: Series of observed output x = {x_1,x_2,………} drawn from an output alphabet V= {𝑣1, 𝑣2, . . , 𝑣_|𝑣|} where x_i belongs to V.

Hidden Markov Model Assumptions

A hidden Markov model is built on several assumptions, including:

1. Output Independence Assumption

Output observation is conditionally independent of all other hidden states and all other observations when given the current hidden state.

Eq. 5: Output independence assumption. | Image: Vivek Vinushanth Christopher

2. Emission Probability Matrix

Probability of hidden state generating output v_i given that state at the corresponding time was s_j.

Markov Model Assumptions

Markov models are developed based on two assumptions:

1. Limited Horizon Assumption

Probability of being in a state at a time t depend only on the state at the time (t-1).

Equation 1: Limited horizon assumption. | Image: Vivek Vinushanth Christopher

That means state at time t represents enough summary of the past to reasonably predict the future. This assumption is an order-1 Markov process. An order-k Markov process assumes conditional independence of state z_t from the states that are k + 1-time steps before it.

2. Stationary Process Assumption

Conditional (probability) distribution over the next state, given the current state, doesn’t change over time.

Equation 2: Stationary process assumption. | Image: Vivek Vinushanth Christopher

That means states keep on changing over time but the underlying process is stationary.

Notation Convention

  • There is an initial state and an initial observation z_0 = s_0
  • s_0: Initial probability distribution over states at time 0.
  • Initial state probability: (π)
  • At t=1, probability of seeing first real state z_1 is p(z_1/z_0).
  • Since z0 = s0:

Notation convention. | Image: Vivek Vinushanth Christopher

State Transition Matrix

State transition matrix. | Image: Vivek Vinushanth Christopher

𝐀𝐢,𝐣: probability of transitioning from state i to state j at any time t.

The following chart is a state transition matrix of four states, including the initial state:

Fig. 2: State transition matrix chart. | Image: Vivek Vinushanth Christopher

2 Questions Answered in a Markov Model

  1. What is the probability of particular sequences of state z?
  2. How do we estimate the parameter of state transition matrix A to maximize the likelihood of the observed sequence?

Probability of Particular Sequences in a Markov Model

Eq.4: Finding probability of particular sequence. | Image: Vivek Vinushanth Christopher

Consider the state transition matrix above. Let’s determine the probability of sequence:

{z1 = s_hot , z2 = s_cold , z3 = s_rain , z4 = s_rain , z5 = s_cold}

P(z) = P(s_hot|s_0 ) P(s_cold|s_hot) P(s_rain|s_cold) P(s_rain|s_rain) P(s_cold|s_rain)

= 0.33 x 0.1 x 0.2 x 0.7 x 0.2 = 0.000924

Hidden Markov Model Example

Consider the example given below, which elaborates how a person feels in different climates.

Fig.3: Markov model as finite state machine. | Image: Vivek Vinushanth Christopher 

  • Set of states: (S) = {Happy, Grumpy}
  • Set of hidden states: (Q) = {Sunny , Rainy}
  • State series over time: = z∈ S_T
  • Observed states for four day: = {z1=Happy, z2= Grumpy, z3=Grumpy, z4=Happy}

The feeling that you understand from a person emoting is called the observations, since you observe them. The weather that influences the feeling of a person is called the hidden state, since you can’t observe it.

Emission Probabilities

In the above example, feelings (“Happy” or “Grumpy”) can be only observed. A person can observe that a person has an 80 percent chance to be “happy” given that the climate at the particular point of observation is sunny. Similarly there’s a 60 percent chance of a person being “grumpy” given that the climate is rainy. The 80 percent and 60 percent are emission probabilities since they deal with observations.

Transition Probabilities

When we consider the climates (hidden states) that influence the observations, there are correlations between consecutive days being sunny or alternate days being rainy. There is an 80 percent chance for the Sunny climate to be in successive days, whereas there’s a 60 percent chance for it to be rainy on consecutive days. The probabilities that explain the transition to/from hidden states are transition probabilities.

How Does a Hidden Markov Model Work?

A hidden Markov model answers three primary questions:

  1. What is the probability of an observed sequence?
  2. What is the most likely series of states to generate an observed sequence?
  3. How can we learn the values for the HMMs parameters A and B given some data?

Probability of Observed Sequence

We have to add up the likelihood of the data x given every possible series of hidden states. This will lead to a complexity of O(|S|)^T. Hence, two alternate procedures were introduced to find the probability of an observed sequence.

Forward Procedure

Calculate the total probability of all the observations (from t_1) up to time t:

𝛼_𝑖 (𝑡) = 𝑃(𝑥_1 , 𝑥_2 , … , 𝑥_𝑡, 𝑧_𝑡 = 𝑠_𝑖; 𝐴, 𝐵)

Backward Procedure

Similarly calculate total probability of all the observations from final time (T) to t:

𝛽_i (t) = P(x_T , x_T-1 , …, x_t+1 , z_t= s_i ; A, B)

A tutorial on how the hidden markov model works. | Video: ritvikmath

Hidden Markov Model Using Forward Procedure

Below is an example of a hidden Markov model using forward procedure.

  • S = {hot,cold}
  • v = {v1=1 ice cream ,v2=2 ice cream, v3=3 ice cream}, where V is the Number of ice creams consumed in a day.
  • Example Sequence: = {x1=v2,x2=v3,x3=v1,x4=v2}

Fig.4: Given data as matrices. | Image: Vivek Vinushanth Christopher

Generated finite state machines for HMM. | Image: Vivek Vinushanth Christopher

We first need to calculate the prior probabilities, that is, the probability of being hot or cold previous to any actual observation. This can be obtained from S_0 or π. From Fig.4, S_0 is provided as 0.6 and 0.4, which are the prior probabilities. Then based on Markov and HMM assumptions, we follow the steps in the figures below to calculate the probability of a given sequence.

1. First Observed Output x1=v2

Fig. 6: Step 1 of the HMM. | Image: Vivek Vinushanth Christopher

2. Observed Output x2=v3

Fig. 7: Step 2 of HMM illustrated. | Image: Vivek Vinushanth Christopher

3. Observed Output x3 and x4

Similarly for x3=v1 and x4=v2, we have to simply multiply the paths that lead to v1 and v2.

Fig. 8: Step 3 and 4 of HMM. | Image: Vivek Vinushanth Christopher

4. Maximum Likelihood Assignment

For a given observed sequence of outputs 𝑥 𝜖 𝑉_𝑇, we intend to find the most likely series of states 𝑧 𝜖 𝑆_𝑇. We can understand this with an example found below.

Fig.9: Data for example two. | Image: Vivek Vinushanth Christopher

Fig.10: Markov model as a finite state machine from Fig.9. data. | Image: Vivek Vinushanth Christopher

The Viterbi algorithm is a dynamic programming algorithm similar to the forward procedure which is often used to find maximum likelihood. Instead of tracking the total probability of generating the observations, it tracks the maximum probability and the corresponding state sequence.

Consider the sequence of emotions: H,H,G,G,G,H for six consecutive days. Using the Viterbi algorithm we will find out the more likelihood of the series.

Fig.11: The Viterbi algorithm requires to choose the best path. | Image: Vivek Vinushanth Christopher

There will be several paths that will lead to sunny on Saturday, and many paths that lead to rainy on Saturday. Here, we intend to identify the best path to sunny or rainy Saturday and multiply with the transition emission probability of “Happy,” since Saturday makes the person feel “Happy.”

Let’s consider a sunny Saturday. The previous day, Friday, can be sunny or rainy. Then we need to know the best path up to Friday, and then multiply with emission probabilities that lead to a grumpy feeling. Iteratively, we need to figure out the best path at each day ending up in more likelihood of the series of days.

Fig.12: Step 1. | Image: Vivek Vinushanth Christopher

Fig. 13: Step 2. | Image: Vivek Vinushanth Christopher

Fig.14. Iterate the algorithm to choose the best path. | Image: Vivek Vinushanth Christopher

 The algorithm leaves you with maximum likelihood values and we now can produce the sequence with a maximum likelihood for a given output sequence.

5. Learn the Values for the HMMs Parameters A and B

Learning in HMMs involves estimating the state transition probabilities A and the output emission probabilities B that make an observed sequence most likely. Expectation-Maximization algorithms are used for this purpose. A commonly-used algorithm known as the Baum-Welch algorithm falls under this category and uses the forward algorithm.

Frequently Asked Questions

What is a hidden Markov model?

A hidden Markov model is a statistical model in which the system being modeled is assumed to be a Markov process with unobserved (hidden) states. It’s used when you can’t observe the states themselves but only the result of a probability function of the states

What’s the difference between a hidden Markov model vs. a Markov model?

    • A hidden Markov model is a probabilistic model used when the results are the product of one of several hidden internal states. 
    • A Markov model is a probabilistic model used to predict a sequence of events when the internal states can be observed.   

What are the basic problems of hidden Markov model?

The three basic problems of a hidden Markov model (HMM) include:

    1. Evaluation or Likelihood Problem: Given an HMM λ = (A, B) and an observation sequence of O = O_1 O_2, … O_T, determine the likelihood P(O'λ).
    2. Decoding Problem: Given an HMM λ = (A, B) and an observation sequence of O = O_1 O_2, … O_T, determine the optimal hidden state sequence of Q (set of finite states).
    3. Learning Problem: Given an observation sequence of O = O_1 O_2, … O_T, learn and adjust the HMM parameters A and B to maximize the probability P(O'λ).

What is Thermodynamic Computing and how does it help AI development?!

Mike's Notes

The reasoning behind this chip is the same as behind Pipi 9. Pipi 9 runs on noise.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library >
  • Home > Handbook > 

Last Updated

17/06/2025

What is Thermodynamic Computing and how does it help AI development?!

By: Laszlo Fazekas
Medium: 05/04/2024

The foundation of modern computing is the transistor, a miniature electronic switch from which logic gates can be constructed, creating complex digital circuits like CPUs or GPUs. With the advancement of technology, transistors have become progressively smaller. According to Moore’s Law, the number of transistors in integrated circuits approximately doubles every 2 years. This exponential growth has enabled the exponential development of computing technology. However, there is a limit to how much the size of transistors can be reduced; we will soon reach a threshold below which transistors cannot function. Moreover, the advancement of AI has made the need for increased computational capacity more critical than ever before.


Transistor count per year from https://en.wikipedia.org/wiki/Moore%27s_law

The fundamental issue is that nature is stochastic (unpredictable). And here, I’m not just referring to quantum mechanical effects. Environmental influences, thermal noise, and other disruptive factors must be considered when designing a circuit. For a transistor, the expectation is that it operates deterministically (predictably). If I run an algorithm 100 times in succession, I must get the same result every time. Currently, transistors are large enough that these factors do not interfere with their operation, but as their size is reduced, these issues will become increasingly relevant. So, what direction can technology take from here? The “usual” answer: quantum computers.

An image of a quantum computer from https://www.flickr.com/photos/ibm_research_zurich/50252942522

In fact, with quantum computers, we encounter the same issue: the need to eliminate environmental effects and thermal noise. This is why quantum computers must be cooled to temperatures near absolute zero. These extreme conditions preclude quantum processors from replacing today’s CPUs. But what could be the solution? It appears that to move forward, we must abandon our deterministic computers and embrace the stochastic nature of the world. This idea is not new. It’s several billion years old.

Educational videos often depict the functioning of cells as little factories, where everything operates with the precision of clockwork. Enzymes, like tiny robots, cut up DNA, to which amino acids attach, leading to the production of proteins. These proteins neatly interlock and, during cell division, separate from the old cell to form a new one. However, this is a highly simplified model. In reality, particles move entirely at random, and when the right components happen to come together, they bind. While human-made structures operate under strict rules, here processes form spontaneously under the compelling influence of physical and chemical laws. Of course, from a bird’s-eye view, the system might appear to function with the precision of a clockwork.

DNA replication from https://en.wikipedia.org/wiki/DNA

A very simple example is when we mix cold water with hot water. It would be impossible to track the random motion of each particle. Some particles move faster, while others move slower. Occasionally, particles collide and exchange energy. The system is entirely chaotic, requiring immense computational capacity to simulate. Despite this, we can accurately predict that after a short period, the water will reach a uniform temperature. This is also a simple self-organizing system that is very complex at the particle level, yet entirely predictable due to the laws of physics and the rules of statistics. Similarly, cell division becomes predictable as a result of complex chemical processes and random motion. Of course, errors can occur. The DNA may not copy correctly, mutations may develop, or other errors may occur. That’s why the system is highly redundant. Several processes will destroy the cell in case of an error (apoptosis), thus preventing faulty units from causing problems (or only very rarely, which is how diseases like cancer can develop).

The energy consumption of a transistor can be comparable to the energy consumption of a cell, even though a cell is orders of magnitude more complex. Imagine the complex calculations we could perform with such low consumption if we carried them out in an analog manner, exploiting the laws of nature.

In biology, thermal noise is not only not a problem, but it is necessary. Below certain temperatures, biological systems are incapable of functioning. It is the random motion induced by heat that powers them.

The foundation of thermodynamic computing is similar. Instead of trying to eliminate the stochastic nature of physical processes, we utilize it. But what can be done with a computer whose operation is non-deterministic?

In fact, in the field of machine learning, there are many random components. For example, in the case of a neural network, the initial weights are randomly initialized. The dropout layer, which eliminates overfitting, also randomly discards inputs. But at a higher level, for instance, diffusion models also use random noise for their operation. In the case of Midjourney, for example, the model was trained to generate images from random noise, taking into account the given instructions.

Here, a bit of noise is added to the image at every step until the entire image becomes noise. The neural network is then trained to reverse this process, that is, to generate an image from noise based on the given text. If the system is trained with enough images and text, it will be capable of generating images from random noise based on text. This is how Midjourney operates.


Steps of Stable Diffusion from https://en.wikipedia.org/wiki/Stable_Diffusion

In current systems, we eliminate the random thermal noise to obtain deterministic transistors, and then on these deterministic transistors, we simulate randomness, which is necessary for the operation of neural networks. Instead of simulation, why not leverage nature’s randomness? The idea is similar to that of any analog computer. Instead of digitally simulating a given process, we should utilize the opportunities provided by nature and run it in an analog manner.

The startup Extrophic is working on the development of such a chip. Like Google, the company was founded by two guys: Guillaume Verdon and Trevor McCourt. Both worked in the field of quantum computing before founding the company, and their chip lies somewhere halfway between traditional integrated circuits and quantum computers.

Extropic’s circuit works in an analog manner. The starting state is completely random, normally distributed thermal noise. Through programming the circuit, this noise can be modified within each component. Instead of transistors, analog weights take their place, which are noisy, but the outcome can be determined through statistical analysis of the output. The guys call this probabilistic computing.

Microscope image of an Extropic chip from https://www.extropic.ai/future

These analog circuits are much faster and consume much less energy, and since the thermal noise is not only non-disruptive but an essential component of the operation, they do not require the special conditions needed by quantum computers. The chips can be manufactured with existing production technology, so they could enter the commercial market within a few years.

As we have seen from the above, Extropic’s technology is very promising. However, what personally piqued my interest is that it is more biologically plausible. Of course, I don’t think that the neurons in artificial neural networks have anything to do with human brain neurons. These are two very different systems. However, the human brain does not learn through gradient descent. Biological learning is something entirely different, and randomness certainly plays a significant role in it.

As I mentioned, in biology and nature, everything operates randomly. What we see as deterministic at a high level is just what statistically stands out from many random events. This is how, for example, many living beings (including us humans) came to be through completely random evolution yet are built with almost engineering precision. I suspect that the human brain operates in a similar way to evolution. A multitude of random events within a suitably directed system, which we perceive from the outside as consistent thinking. This is why genetic algorithms were so intriguing to me, and now I see the same principle in Extropic’s chip.

If you are interested, check the company homepage or this interview with the founder guys.

Lindy Effect

Mike's Notes

This is a curious thing.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library >
  • Home > Handbook > 

Last Updated

17/05/2025

Lindy Effect

By: 
Wikipedia: 2024

"The Lindy effect (also known as Lindy's Law) is a theorized phenomenon by which the future life expectancy of some non-perishable things, like a technology or an idea, is proportional to their current age. Thus, the Lindy effect proposes the longer a period something has survived to exist or be used in the present, the longer its remaining life expectancy. Longevity implies a resistance to change, obsolescence, or competition, and greater odds of continued existence into the future.[2] Where the Lindy effect applies, mortality rate decreases with time. Mathematically, the Lindy effect corresponds to lifetimes following a Pareto probability distribution.

The concept is named after Lindy's delicatessen in New York City, where the concept was informally theorized by comedians. The Lindy effect has subsequently been theorized by mathematicians and statisticians. Nassim Nicholas Taleb has expressed the Lindy effect in terms of "distance from an absorbing barrier".

The Lindy effect applies to "non-perishable" items, those that do not have an "unavoidable expiration date". For example, human beings are perishable: the life expectancy at birth in developed countries is about 80 years. So the Lindy effect does not apply to individual human lifespan: all else being equal, it is less likely for a 10-year-old human to die within the next year than for a 100-year-old, while the Lindy effect would predict the opposite. ..." - Wikipedia