Zeit·geist = spirit, essence of a particular time

A collection of food-for-thought posts and articles on technology, business, leadership and management. 

Race of the Rest: The Unicorn Trend has gone Global

Love this pic, had to post it again, makes a perfect background for this topic.

It was Steve Case, former AOL, who has championed the term 'Rise of the Rest' on the emergence of tech hubs and VC out of traditional coastal zones in US.

This is not a US domestic phenomenon, the Unicorn trend (startups valued over $1 Bn) has gone global after being in the works for a few years now. See here for educated predictions back in 2015 (somehow neglected at that time, now endorsed by stats).

Thanks to CB insights who just released their Christmas newsletter yesterday, we have some closing year stats to share:

2016 saw close to 300 deals of VC-backed companies in overlooked regions such asLatin America, Africa and Oceania scoring record numbers in terms of # of startup deals.
In terms of overall investment in these regions, this year ended with $1.3 Bn, with three years in a row now consistentlyover $1Bn in capital deployed.

More interesting even, and, as anticipated almost two years ago, flashback here:

of the 1.063 deals done in these three regions since 2009, Latin America alone took almost half of them or 47%.

The world is truly changing.

Happy Holidays


Machine Learning: an industry perspective

by Garret Robertson - Senior Analyst & Author


Satyajeet Salvi
Ruben Ramirez
Ellen Chan

The future of technology is in machine learning. Talk of virtual assistants, neural networks and deep learning is proliferating across the Internet at a rapid pace. According to a recent CB Insights update, deal flow in this space is accelerating rapidly with current estimates of the industry size exceeding $100 billion with compounded annual growth estimated at over 50%. Despite the proliferation of this technology, it is misunderstood. Dreams of androids, self-driving cars and Skynet abound in the conversations of executives, the general population and everyone in between.

Machine Learning industry size is exceeding $100 billion with compounded annual growth estimated at over 50%

Machine learning tools can more accurately described as powerful tools that sort through terabytes of data in order to optimize relationships. These tools find solutions for minimizing fraud, maximizing sales revenue, maximizing lead generation, or minimizing errors in image recognition. What makes these algorithms truly special is their ability to take complex structured and/or unstructured data and find meaningful relationships. Some examples of these data sources can include text heavy sources such as emails or web sites, images, audio files, and/or data points.

Despite lack of awareness, these tools are already finding places in consumers’ lives. When consumers log into Netflix at night and pick a show from the recommended play list or when they choose to add a recommended product to their basket on Amazon, machine-learning algorithms are at the heart of those lists. It is not just limited to product recommendation though, when consumers’ credit cards deactivate over suspicious transactions, there was machine learning. When Social media presents ads to users, there again was machine learning. Additionally these powerful algorithms drive other services like the virtual assistants Siri, Cortana or Alexa. While these examples may be visible to consumers, Machine Learning is rapidly proliferating into many less visible markets like CRM, healthcare and government services and banking.

The valuation of machine learning service companies can best be described by its synergies with cloud service providers and businesses. Businesses create systems that gather data as they conduct business. These systems could include, as an example, systems for tracking customer receipts like an accounting ledger or a customer-profiling tool like a rewards program. Data science showed businesses how to combine these two data sets to better understand customer preferences. When the data moved to cloud services, machine-learning tools were then able to sift through much more complicated information like images, articles, or other unstructured sources and automate the search for interesting relationships.

The valuation of machine learning service companies can best be described by its synergies with cloud service providers

As the outputs became better, the businesses rebuilt systems to integrate more data necessitating more data storage. Now the systems could create profiles, link them to purchasing trends and compare it to even more complex demographic information creating more powerful business insights. The outputs from the 3-way cycle thus reinforce themselves making it more and more efficient and increasing value to all parties.

These synergies define how the industry has been growing. Because the synergies are so strong, most capital investments in this industry occur as partnerships between businesses, cloud service providers and machine learning companies. These strategic investment partners provide two critical pieces to the growth round. First, they validate the effectiveness of the machine-learning product. Second, these partnerships provide access to data from interesting industries such as fraud, healthcare, product recommendation or sales analytics allowing opportunity for the systems to become even more effective.

most capital investments in this industry occur as partnerships between businesses, cloud service providers and machine learning companies

Below is a sampling of some capital raises for machine learning companies where at least one of the investors was not a capital player but a business with a strategic interest and/or a cloud service provider.

The application of this technology is expanding every day. Nearly 70% of all investment into this space is driven by Seed and Series A funding. Additionally more than 40% of all companies that exist in this space are less than 3 years old. Additionally, with the power these solutions have to offer, the industry is expanding rapidly with total year over year transaction and investment volume increasing.

Nearly 70% of all investment into this space is driven by Seed and Series A funding
More than 40% of all companies that exist in this space are less than 3 years old

Due to the synergies in this industry, a few companies have been able to lead the charge. Some of these companies include Amazon, Microsoft, Google, IBM and Apple. This makes sense because the effectiveness of the algorithms grows as the access to relevant data grows. Companies with access to large quantities of data find more value than those with less.

Despite the power of machine learning, there remain two important hurdles for the typical company in adopting these technologies. First, company leadership needs to be aware of how these systems can help them. Understanding how data can be used to redefine and refine existing strategies is crucial in transforming the organization’s systems. General misunderstanding of machine learning has prevented many companies from adopting it.

The effectiveness of the algorithms grows as the access to relevant data grows

Second, if companies want to pursue implementation of these systems, they need to understand how. This involves not only utilizing tools to gather the data, but also knowing what kinds of solutions are already available.

There are many machine learning companies such as BigML, Amazon, IBM, Microsoft, Google or others that have out-of-the-box solutions available to a wide range of industries. Increasingly, machine learning is moving from the world of PhD’s and large teams of data scientists to tools that anyone can implement.

Despite the newness of this technology to businesses, many industries have already found interesting and powerful solutions. A summary of some industries that have been impacted by machine learning as well as some specific examples in selected industries follows.


This is one area where the use of machine learning is most visible to consumers. When customers buy products online, they leave behind with the business a treasure trove of information. Some of this information includes what products are typically bought together, how much the average consumer spends in a given purchase, what sorts of products and brands people like and much more. While individual tickets report single transaction information, registered users create entire shopping profiles over multiple purchases that can be analyzed.

With this kind of information, it is no wonder that Amazon reported shortly after rolling out its product recommendation platform that sales increased by nearly 30%. In fact this is not an uncommon story. With companies better able to identify the needs and wants of users, they are better able to put products consumers want, into their hands.

In addition to product recommendation, chatbots are taking over customer service. More than 11,000 bots have been added to Facebook Messenger since its launch, allowing brands and companies to use AI to connect with customers through virtual concierge services. These bots are replacing employees in physical stores, allowing companies to build long-term relationships with customers while saving labor costs.

Spring Bot is one example of many of these services that acts as a point of contact even after purchases are made and has a wide range of customers, including Givenchy and Lanvin, brands that do not have an established e-commerce platform. An automated interaction generally costs $0.25, while a live agent interaction costs anywhere from $6 to $20. The automated interactions are also faster than normal live interactions. While the natural language processing in these systems is not perfect, the overall results speak for themselves.


Increasingly machine learning tools are being used to enhance sales and CRM. Traditionally, sales data has been stored and analyzed manually. In addition to the time and money spent in performing these tasks, significant capital has been spent training sales teams to track the right data and how to effectively analyze it.

Machine learning has provided a way to collect data automatically and provide the analysis so sales agents can more effectively find, target and convert prospective clients into sales. InsideSales reports that some of its customers have increased their sales pipeline by 30% increase to sales and a 250% increase in leads. Costs associated with training implementation and data entry are reduced for users in addition to these strong revenue increases.

Financial Services:

Increasingly, financial institutions are using automated financial advisors and planners. These tools monitor events and stock and bond price trends and compare them to the user’s financial goals. The machine will be able to compare the user’s portfolio and make recommendations on what stocks to buy or sell. There will be no need to pay an expensive human advisor to make decisions for customers. The machine-learning tool will now be able to make decisions based on data that is coming in real time.

In addition to automated financial advisors, algorithmic trading is a means to increase profitability and decrease risk in investment portfolios. Algorithmic trading systems are systems that process data on a very large scale to identify risks in investment portfolios and rebalance them in order to minimize risk. As these systems gain more data, they are better able to optimize portfolios and mitigate risk.


It is estimated that these algorithmic trading systems handle 75% of the volume of the global trades worldwide. These numbers get larger when looking at specific types of trading.

Algorithmic trading systems were responsible for nearly 80 per cent of foreign exchange futures trading volume, 67 per cent of interest rate futures volume, 62 per cent of equity futures volume, 47 per cent of metals and energy futures volume, and 38 per cent of agricultural product futures volume between October 2012 and October 2014.


Clinical variation management is an area in healthcare ripe for disruption by ML systems. Clinical variation is when clinicians deviate from recommended care pathways in the delivery of care to similar patients. It is estimated that there could be as much as 30% waste in healthcare, but this waste is hard to identify due to the complexity of healthcare and the great degree of variation in the way patients receive treatments.

A recent article by HealthCatalyst indicates the problem. Clinical variation is complicated by two main factors. The first is that studies indicate that only 20% of the care delivery is driven by scientific research. About 80% of the care delivery is determined by subjective clinical care pathway decisions. Second, Doctors must read hundreds of pages of primary literature every day in order to stay fully current. The process needs education, but the education is next to impossible to get and train through normal means. Until then, clinicians deliver care without much consistency driving waste and impeding process development.

Machine learning provides a means to monitor care pathways to ensure clinical variation is minimized. It also provides a means to monitor care pathways to determine areas to improve and optimize them with current methods in mind.

Traditional tools such as control charts, regressions and manually examined data are not robust enough to optimize the system. Machine learning tools are well positioned to do the work individual data scientists and analysts cannot do. Those machine-learning companies focused on healthcare like Ayasdi are well positioned to disrupt this space.

Concluding Remarks

  • Machine learning as an industry is still in its infancy.
  • These examples represent only a few of the hundreds of companies that are emerging to solve next generation business problems.
  • A new industrial revolution is coming in the form of computer code and automated data science.
  • Companies that are not thinking about data and machine learning will soon find themselves unprepared.
  • The companies who have adopted these technologies already enjoy significant advantages over those who have not implemented it yet. **

*special note of thanks to Naiss' contributors:

Satyajeet Salvi
Ruben Ramirez
Ellen Chan













Let’s Get Paid Upfront: getting over the carrot-stick game

In start-up and VC land is all about technology innovation, isn’t it?… or so we tend to think.

Actually is not, or at least not all, many companies are thriving on simple innovation schemes applied to processes, business models or simple tactics.

Going further, the concept itself of ‘frugal innovation’, a.k.a creative thinking in the face of constraints, a.k.a doing more with less in the face of critical conditions, shows all the many ways problems can be solved even in the absence of proper resources.

This wonderful TED talk by Navi Radjou is an excellent primer on the topic, covers from fridges running with no electricity in India to advertising billboards producing water literally out of thin air in the rain scarce city of Lima in South America.

But today, in this post, we’re going to talk about an even more simplistic yet powerful innovation tactic: incentives.

First, the Theory:

Incentives and reward schemes are used in companies to drive employee and management behaviors, aligning those toward a set of objectives.

Incentives range from simple weekly, monthly or yearly salaries all the way to sophisticated cash bonuses with triggers, accelerators and equity schemes more long term oriented.

For the vast array of different incentive and compensation mechanisms one thing is true, they all are paid AFTER the activity and expected behavior has happened. Simply put, if you behave and meet your targets, you get paid (carrot), if not, you don’t (stick).

…There must be a better way. Considering everybody needs to be paid for his/her work, what if we get paid BEFORE?.

With current company incentives lying ahead, in the future, all rewards are perceived as a future ‘gain’ contingent on employee & management behavior and this is an excellent transaction scheme, no doubt.
But there is a much bigger motivation factor for humans, actually double at least: loss aversion.

Kahneman, D. and Tversky, A. (1984) — ‘Choices, Values and Frames’

Kahneman, D. and Tversky, A. (1984) — ‘Choices, Values and Frames’

Just look at the grey area in the left (loss quadrant) in comparison with the green area on the right (gain), for the same absolute incentive amount X, you get 2Y motivation for a loss.

So, what if we re-wire the incentive schemes in a way we can leverage this much bigger motivating factor?

The Watney rule for Startups

In this letter to their limited partners, First Round venture capital firm set the stage for what is to come this year for start ups.

Back to the ‘old normal’ and adjusted valuations has put all expectations on start-ups to become deeply conscious on their expenditure and make the most efficient use of capital.

Kind of the behavior the astronaut Watney showed while stranded on Mars waiting for rescue in the movie The Martian.

Similarly, and while the next ‘rescue’ round of capital comes, Start-ups CEOs need to use their limited resources imaginatively while securing key milestones and monetization in particular happens. So they started paying bonuses in advance.

Wait… how is that an efficient use of working capital?

It actually is, and, as Irfan Pardesi told me when we met in San Francisco in May, paying in advance his sales execs was a much better motivator for performance, which grew 15% on average and even 50% in some individual cases.

Irfan is a serial entrepreneur, founder of Accentuate Capital Markets, a holding company providing financial FX brokerage services from South Africa. Irfan is also a member of YPO, the Young Presidents Organization, a non-profit organization of leaders under 45.

Paying bonuses and/or the paycheck in advance creates trust and strengthen bonds between the company and its employees/managers, who feel much more confident about the company and react incredibly well to the trust put on them.

On top of that, think how cool it is to get your money upfront every month, so you can do the things you want now rather than waiting to get that elusive bonus at the end of the year. This is a form of instant gratification in a way, and your company is doing that for you in appreciation of your work.

There needs to be a catch of course for this model to work and to secure ‘loss aversion’ dynamics work towards objectives.
Irfan told me, for this model to work two things must be in place:

  • Close management and monitoring of the activity so targets are really realistic and achievable, consistently, each month

  • A catch, a future deduction of the paycheck/bonuses if targets are not met, something reasonable and agreed with the employees

End result? once the process is tuned in, and, securing you build that trusted relationship with your employees, productivity and performance will be boosted significantly (specially at the end of the month when loss aversion feelings kick in).

The magic of a 2x motivating factor.

Actually there is a whole new industry arising around the concept of instant gratification, the precursor to loss aversion.

ActiveHours, a Palo Alto startup offers the possibility to get your monthly salary paid almost in real time, through an app and in cozy hourly installments.

It’s like everyday is pay-day!

I’m particularly fond of the trusted relationship model, in my own experience, all stelar teams and epic growth stories come from nimble teams with strong bonds of trust both amongst them and with management, add the right incentives at the right timing, and there you go, the sky is the limit.

ed fernandez

How to make machines learn like humans: Brain-like AI & Machine Learning

AI and machine learning are all over us, a simple search on google draws 105 Million entries and counting, google trends shows a growing demand for this search term consistent with the exponential rise of deep learning since 2013, more or less when Google’s X Lab developed a machine learning algorithm able to autonomously browse YouTube to identify the videos that contained cats.

In 1959, Arthur Samuel defined machine learning as a

Field of study that gives computers the ability to learn without being explicitly programmed

AI and machine learning changes the software paradigm computers have been based on for many decades.

In the traditional computing domain, providing an input, we feed it into an algorithm to produce the desired output. This is the rule-based frameworkthe majority of the systems around us still work with.

We set up our thermostat to a desire temperature (input) and a rule based programming (algorithm) will take care of reading a sensor and activating heating or AC machines to get to the room temperature we want (output).

The industry has been working relentlessly for many years developing better hardware, software and apps to solve a gazillion problems and use cases around us with programmable solutions. But still, every new functionality or feature, every single new ‘learning’ has still to come via an update of the software (or the firmware itself in hardware).

> Machine learning puts head over heels the rule-based paradigm.

Given a dataset (input) and a known expected set of outcomes (output), machine learning will figure out the optimal matching algorithm so that, after trained (learning), it can autonomously predict the output corresponding to new inputs.

The new AI and machine learning paradigm opens up the promise land of ‘self-programming’ machines, capable of finding the right algorithms to be used in any occasion, this is, providing availability of sufficient input training data, the bottleneck today.

However, and despite all the incredible progress made in this field, including breakthroughs around deep learning in recent years, machines are far from matching human ability to learn new patterns, and worse, we don’t know how they learned what they learned nor how they come up with a decision or an specific (wrong) output. We just feed them with big data and ‘tweak’ the machine learning process till we get them to work and deliver the desired outputs within acceptable thresholds of accuracy, but the whole thing remains a ‘black box’ (Fodor & Pylyshyn 1988, Sun 2002).

And they got very efficient and accurate, better than humans in many fronts, no question. AI, machine learning and neural networks are now behind any major service, predicting our credit score, detecting fraud, rendering face recognition, assisting us through Siri, Google, Cortana or Alexa, and soon driving our cars.

But, as in the old computing paradigm, the process of learning still requires an ‘update’, what, in machine learning and neural networks jargon is called ‘retraining the network’ with a new dataset and new features required to incorporate a new learning or a new functionality.

Retraining any AI network takes well experience engineers, top notch hardware (GPUs) and time, a lot of computing time.

That’s why we can’t teach Siri, Google, Cortana or Alexa new things on the fly. If they don’t understand what we say they typically default to a simple search on the web, we can’t simply tell them ‘learn this new word’ or ‘remember my favorite team are the Red Sox’. Same applies for the rest of large neural networks behind other services, they need to be retrained with the new data and that takes days, weeks or even months depending on the size of the the network and the dataset.

Now, imagine for a moment if we could teach machines ourselves and make them learn the same way we humans do, wouldn’t that be awesome?
Imagine if we could teach Siri, Cortana, Google o Alexa new words or expressions, or even new action commands ‘hey Alexa, pull out my car from the garage’:

The answer to this is in the brain.

And some researchers, devoted to reverse engineer the recognition mechanisms of the brain have unlocked brain-like algorithms and new machine learning models solving this problem, turning the traditional machine learning ‘black box’ into a ‘clear box’ neural network where new learnings can happen on the fly, in real time and at a fraction of today’s computational cost (no retraining over the whole dataset required).

In a simplistic way, the underlying problem is that all traditional machine learning models are primarily feedforward based, in other words, the basic calculations in the network happen ultimately in the form of a simple multiplication where the output Y is just the input X weighted (feedforward multiplied by W, the Weight). Y = W * X

Determining the set of weights W for a given input dataset X with a known labeled output Y is called ‘training the network’. The process is long, can take hours, days or even months for large networks, but, once all those weights W are calculated and refined (a process called optimization of weights) the network is capable of amazing wonders like facial recognition or natural language understanding.

However, as mentioned, if you want the network to learn something new, you need to go back again through all the retraining process and start from the beginning, recalculating and optimizing the new set of weights W.

But ‘this is not how the brain works’, Tsvi Achler, MD/PhD in neuroscience and BSEE from Berkeley, told us at a talk in Mountain View.

‘The brain does not turn around and recalculates weights, it computes and learns differently during recognition while the context is still available, and it does not only do feedforward, all sensorial neural recognition mechanisms show some form of feedback loop’
In all traditional machine learning methods (deep learning, convolutional networks, recurrent networks, support vector machines, perceptrons, etc) there is a ‘disconnect’ between the learning & training phase and the recognition phase. What Tsvi Achler proposes is not to recalculate (learn) weights but to determine neural network activation (Y, output) by optimizing during recognition, factoring in feedback as well as feedforward, and more importantly, focusing on the current pattern in context (vs all of the training dataset).

With this approach and this new machine learning algorithm we can ‘see’ the weights and change them in real time, while recognizing, add new nodes to the network (patterns) and features on the fly without the need to go over the re-training process.

At his startup Optimizing Mind, Tsvi ran his machine learning model on a Celeron Quad core laptop, 2GHz, 4GB memory, which is equivalent to a high end smartphone. He tested it against traditional methods such as SVM or KNN and the scalability results were astonishing, showing off up to two orders of magnitud of computational cost reduction.

The ability to embed this new machine learning technology in a smartphone will enable true real time learning from end users’ interaction while preserving data locally (no need to go back and forth to servers).

The time when we will be able finally to teach machines by ourselves, as well as learn from the environment, all in real time, is getting closer.

This may be even a first very early step to enable machine to machine learning, and with that, who knows, exponential intelligence maybe?.

Exciting times ahead, what a moment to be alive.


Two Duck-Rabbit Paradigm-Shift Anomalies in Physics and One in Machine Learning

You never know what a meeting for a quick coffee in Palo Alto can turn into.

What was supposed to be an ‘informal’ chat (if there is such thing when talking with PhD’s) about feedforward-feedback machine learning models, turned into a philosophical discussion on duck-rabbit paradigm shifts.

(disclaimer 1: I’m just a nerd without credentials either topic you choose, with a genuine interest though)

First, the theory:

I see a Rabbit, You see a Duck

Thomas Kuhn described the nature of scientific revolutions back in 1962 (his book The Structure of Scientific Revolutions).

A contrarian back in the time, as he re-defined progress by moving from development-by-accumulation (on pre-established assumptions) into paradigm shifts, or revolutions in scientific progress by looking into anomalies inferring a drastic change of assumptions.

In other words, Kuhn advocated for a change of rules over the pre-existing framework as the ultimate scientific progression method.

The Copernican revolution, Newton’s reformulation of gravity, Einstein’s relativity or Darwin’s evolution all were ‘anomalies’ as theories.

Sun vs Earth at the center, relativity vs linear spacetime, Apes evolving into Humans vs creation

The ethos of the scientific progress theory rests on identifying the right anomalies which support new paradigms. Anomalies come up as revolutions in disguise and, utterly (> and I love this> ), expand on the previous paradigm which ends up nested within remaining perfectly valid.

Anomalies create rejection by opposition (it’s a Duck!, no is not, it’s a Rabbit), but after the new paradigm takes over (…I can see the Duck now ?!?) both paradigms co-exist (it’s a duck AND a rabbit!, illustration above)

For a true paradigm shift to happen, the anomaly needs to grow from exception into alternative: ‘it’s a Rabbit AND a Duck!’

Ok, fair enough on the lecture, where is this Machine Learning anomaly this click-bait headline was all about?

It’s coming, bear with me, will be worth the reading while we get there, first, a couple of jaw-dropping-no-longer-anomalies-but-paradigm-shifts: the first one explains the origin (and meaning) of life, the second one may redefine physics forever.

1. Dissipation-driven adaptation

Jeremy England. MIT, biophysics

This incredibly simple idea is intuitively so powerful and makes so much sense that is difficult to resist. It explains Darwinian evolution and survival of the fittest, ultimately dwelling on the inherent reasons why life comes to exist.

At an intuition level, in Jeremy’s words:

You start with a random clump of atoms, and if you shine light on it for long enough, it should not be so surprising that you get a plant
— Jeremy England

Jeremy, a MIT’s researcher, has developed a mathematical model based on current physics, exerting that a given set of atoms, exposed to a continuous source of energy (i.e the Sun), surrounded by a hot bath (i.e the Ocean) will self-organize to dissipate energy in the most efficient way (i.e life).

We, carbon-based lifeforms, in Spock’s vulcan language, are much better at dissipating heat than inanimate objects. Both living and non-living organisms show this efficiency driven, self-organizing dissipation behavior.

Photosynthesis and self-replication (of RNA molecules, precursor to DNA-based life) are consequences of dissipation driven adaptation. Photosynthesis is about capturing sunlight energy transforming and storing it chemically (sugar) so it can be transported and reprocessed for plant growth and replication (hence forests).

Don’t believe it yet? see for yourself, here is Dr. Hubler’s Stanford professor experiment on self-wiring ball bearings, an example of dissipation driven matter structure reorganisation.

2. Timeless physics

Julian Barbour. British Physicist. Quantum gravity.

Remember the school/college days?: Speed = space / time, power = energy / time, theorem of calculus df/dt, Maxwell’s equations, Einstein’s relativity, Thermodynamics, etc, etc. In physics, anything dealing with change, requires t (time) as a variable, isn’t it? …may be not any more.

How is it possible anyone dares to defy physics by removing time from centuries old proven equations?

If you think about it, time is just an abstraction we use to facilitate our understanding of how things (matter in particular) transitions from one state to another (change). Because we live in an universe governed by the 2nd law of thermodynamics (fighting an increasing entropy) we perceive linear time as our most reliable and dependable reference.

At an intuition level, if we look at the Universe as a simple but immense ‘cloud’ of matter in permanent change (motion) since the big bang occurred, then, if we reduce our view to atoms transitioning for one state to another, you could remove time entirely.

Our Universe could be viewed as a continuum of matter in ‘motion’ (actually, according to Barbour, not motion, but matter in permanent change, removing in full the spacetime continuum).

Our senses and limited computing capacities can’t deal with such enormous entity so we take partial ‘pictures’ with a reference point (time) to deal with reality and make sense out of it (a constrained and partial view).

Another intuitive line of thought, if Newton’s physics were based on linear time (absolute fixed time), and then Einstein’s relativity made time relative, hence flexible (unlocking a bigger scope for physics), what if we make time super-mighty-flexible to the point of making it irrelevant? wouldn’t this even offer an even wider and extended view as we remove the constraints of a time dimension itself?

If, at first, the idea is not absurd enough, then there is no hope for it
— Albert Einstein

…. and now, for something completely different (Monty Python)

3. Flexible recognition in machine learning

Tsvi Achler. Neuroscience (PhD), Medicine (MD), Electrical Engineering (BS-EECS Berkeley) — Optimizing Mind

Our brains are ‘computationally flexible’, this means we can immediately learn and use new patterns as we encounter them in the environment.

We actually ‘like’ to develop those patterns, as we unleash our curiosity, see and try new things for the sake of enjoyment.

Learning, tasting and traveling feed our brains with new patterns. Riding a hover-wheel, flying a drone, speaking to Amazon echo or playing a new game are examples of behaviors where our brains confront and develop new patterns for different uses and purposes.

Now, let’s look into it from a machine learning perspective:

(disclaimer 2: as said at the beginning of this post I’m just a nerd without credentials trying to convey the message. Standing in the shoulder of giants when I wrote what you’re about to read on)

Tsvi Achler has been studying the brain from multidisciplinary perspectives looking for a single, compact network with new machine learning algorithms and models who can display brain phenomena as seen in electrode recordings, performing flexible recognition.

The majority of popular models of the brain and algorithms for machine learning remain feedforward and the problem is that even when they are able to recognitze they are not optimal for recall, symbolic reasoning or analysis.

For example you can ask a 4 year old why they recognized something the way they did or what do they expect a bicycle to look like. However it is difficult to do the same with current machine learning algorithms. Let’s take the example of recognizing a bicycle over a dataset of pictures. A bicycle, from a pattern perspective, would consist of two same or similar size wheels, a handle, and some sort of supporting triangular structure.

In feedforward models the weights are optimised for successful recognition over the dataset (of a bicycle in our example). Feedforward methods will learn what is unique within a bicycle compared to all other items in the training set and learn to ignore what is not unique. The problem is that subsequently it is not easy to recall what are the original components (two wheels of same of similar size, a handle, a supporting triangular structure) that may or may not be shared with other items.

Moreover when something new must be learned, feedforward models have to figure out from what is unique to the new item but not to the bicycle and other items it already knows how to recognize. This requires re-doing learning and rehearsing all over the whole dataset.

What Tsvi suggests is to use a feedforward-feedback machine learning model to estimate uniqueness during recognition by performing optimization on the current pattern that is being recognised, and determining neuron activation. (this is NOT optimization to learn weights by the way).

With this distinct model, weights are no longer feedforward, learning is more flexible and can be much faster, as there is no need to do rehearsal over the whole dataset.

In other words, this model is closer to how our brain actually works, as we don’t need to rehearse a whole dataset of samples to recognize new things.

Think about it, how many samples of the much hyped hoverwheels do you need to see first before recognizing the next one on the street?. Same for a bicycle.

And, the most important thing, with feedforward-feedback models learning happens with significantly fewer data.

Much less data required to learn, an much faster learning.

Optimization during recognition displays also properties observed in brain behaviour and cognitive experiments, like predicting, oscillations, initial bursting with unrecognized patterns (followed by a more gradual return to the original activation) and more importantly even, speed-accuracy trade off (so here is your catch if you were looking for it).

All in all, feedforward-feedback models will make machines learn faster using less data.

They also mimic better how our brain works.

I met Tsvi for the first time at a talk in Mountain View: available here. I will be helping him and his startup along his journey which (as all new ventures) starts with funding, so if anyone has an interest or wants to know more please do not hesitate to reach out and leave a message for Tsvi or me in the comments, or even better, tweet me at @efernandez.

Thanks also to Bart Peintner, Co-founder & CTO at Loop.ai, for his advice, insights and shared interest for the ideas mentioned in this article (note-to-ourselves: keep always bandwidth in your mind to entertain challenging ‘anomalies’)