blennon

I highly recommend Harmonious.ai written by the folks at AI2 Incubator who are brilliant and reputable.

This is tangential but does anyone have a "better explanation" for topology?

I did my Ph.D. work on the cerebellum and proposed a theoretical model of reinforcement learning and can add some context for why this is a "big deal".

First, it represents a fundamental shift in how the cerebellum learns. For a long time, it was thought that the cerebellum learned liked a perceptron neural network, i.e. that an error signal was computed and used to change the strength of synapses within the cerebellum to result in the correct output signal to guide motor control (and, we know now, cognitive control). In other words, the cerebellum was a supervised learning machine [1]. But how these error signals were actually computed using the neuronal circuitry was never made clear; most arguments centered around the microcircuitry of the inferior olive. In a perceptron, an error signal is the difference between the correct output and the predicted output. But how is the "correct output" supplied?

As far as I know the first to propose the cerebellum learned by reinforcement learning was the famed cerebellum researcher Richard Thompson [2]. Unfortunately, the idea was only vaguely sketched out the field didn't take this very seriously and continued on with the general belief that the cerebellum learned by supervised learning.

To me and my collaborator, Tadashi Yamazaki, it seemed a more natural signal that the nervous system could capably supply would be a graded reward signal. Moreover, this meant we could interpret the structure of the cerebellum within the theoretical frameworks of reinforcement learning such as the actor-critic framework [3,4]. This is the second reason these finding are a big deal: the paradigm shift to the cerebellum being a reinforcement learning machine, if correct, will be a boon for building better models of it. In the last few years there has been some impressive work done with reinforcement learning in artificial neural networks that could be applied to models of the cerebellum, especially within context of the brain at large.

[1] Doya, Kenji "What are the computations of the cerebellum, the basal ganglia and the cerebral cortex" (1999) [2] Thompson, Richard "The nature of reinforcement in cerebellar learning" (1998). [3] Lennon, William "Towards more biologically plausible computational models of the cerebellum with emphasis on the molecular layer interneurons" (2015) [4] Yamazaki and Lennon "Revisiting a theory of cerebellar cortex" (2019)

Anyone around Pasadena, CA interested in a meetup? Santa Monica is...far.

This is a great example of a niche banking solution which is far better than any traditional bank can provide. I think we'll see a lot more of this as banking APIs become more available.

My company is working on similar problem and solution for youth sports teams and clubs which are also non-profits (formally or informally).

Any idea what they are? We've seen an uptick in users reporting transactional emails going to spam.

Groundwork | Full-Stack Developer | Los Angeles | Onsite or Remote | Full-Time

We’re on a mission to make it easier for youth sports teams and clubs to manage money responsibly and transparently.

Groundwork is developing a financial management platform that makes it easy to budget for the season, collect player dues online, and pay for expenses all with complete transparency to stakeholders. We give volunteer treasurers their time back and give parents visibility into where their money is spent.

We’re looking for a mid-level or senior full-stack engineer to join our team and play a key role in product development. We started 3.5 years ago, are investor-backed, generating revenue and handle significant payment volume every month.

Our stack: Python, Flask, Postgres, HTML, CSS, Javascript

Full job description: https://angel.co/teamgroundwork/jobs/494861-software-enginee...

If you’re interested, we’d love to hear from you: jobs [at] ground.work

Connecting this to the OP is that it looks like the stimulation is initially activating these CPGs -- all of the movement is happening based on control from the spinal cord alone. It's not clear how much communication is taking place between the motor cortex and the spinal cord in this patient.

Something to add to this that most non-neuroscientists don't realize is that the spinal cord is not just a bundle of wires that connect the brain to muscles. The spinal cord is a complex circuit that contains "central pattern generators" (CPG) that can produce rhythmic movement. When activated, these networks can be pushed into a dynamic state where they repeat the same movement repeatedly, e.g. walking. For more, look up the work by Sten Grillner on stingray CPGs.

One approximate way of looking at this is that the brain's motor cortex sends a "go" signal to the spinal cords CPGs and these start generating the signals to the muscles to walk. You can also think of motor control as hierarchical where as you go up the hierarchy. The bottom of the hierarchy are the muscles/actuators, then the neurons that stimulate the muscles, then the circuits in the spinal cord that have these CPGs and other "primitives", then primary motor cortex and further up the cortical hierarchy. At the higher levels of the cortical hierarchy are representations of whole movements, like moving your hand to mouth. As you descend the hierarchy, the neural signals control gradually more details components of the movement. For more, look up the work by Michael Graziano.

Indeed. What's new is what sort of information is conveyed from the gut to the brain via the vagus nerve. Note: information is also conveyed in the other direction from the brain to the gut.

This study shows that your gut is equipped with environmental sensors (enteroendocrine cells) that can convey nutritional information from the gut to the brain on short time scales (~seconds or faster). It's like your gut's "eye" in that it can "see" what's going on, like sugars are present, and convey this to your brain.

I'm not an expert on the gut-brain field but I've always wondered where food cravings come from, especially specific ones that pregnant women often exhibit. E.g. craving dirt. My conclusion is that your body must have a way of sensing what nutrients you're ingesting and missing.

Young, charismatic ivy league grads are one specific instance of "the quality of founders ". Another, equally fitting one is, "A successful serial entrepreneur with an exit or two under their belt". Another, albeit seemingly outdated one is, "a former Googler". There are many ways a founder can be of high quality that are not ivy league grads.

If I could downvote your comment I would because this is not an accurate synthesis of the article.

"more young, charismatic ivy league graduates" is not at all what the author describes. The author's thesis is that raising a Series A has requirements ranging between a Seed round which is based on, "the quality of the founders and the raw story that they can tell about their company and the future that company will create" and a Series B which is based on "[the] need to have accomplished a significant set of things that prove their ability to accomplish that future".

What I find most interesting is the multiple training methods used to get the network to improve its performance. They name a few in the article:

- dual learning - deliberation networks - joint training - agreement regularization

I haven't read the paper to see how these are combined but it makes intuitive sense that using multiple training methods can lead to better performance. That is to say, to more effectively search the weight space of the network.

Can you elaborate on this?

"The result of the computation is also stored in the memory devices, and in this sense the concept is loosely inspired by how the brain computes."

For anyone who is interested in a simple model of how the brain does this, check out "associative memories". The basic idea is that networks of neurons both store memory (in their synapses) and perform the computations to retrieve or recall those memories.

A simple example is the Hopfield network, a single layer of neurons that are recurrently connected with a specific update function: https://en.wikipedia.org/wiki/Hopfield_network

Another is two layers of neurons that are reciprocally connected called a Bidirectional Associative Memory (BAM): https://en.wikipedia.org/wiki/Bidirectional_associative_memo...

edit: grammar

The biggest advance that I've seen towards AGI is the work using reinforcement learning, e.g. neural nets that learn to play video games through trial and error. There is an impressive repertoire of _behavior_ that emerges from these systems. This, in my opinion, has the greatest potential to take us another big step towards -- but not necessarily to -- AGI.

Now that I'm in a position of hiring technical people, this seems crazier than ever to me. It's hard to hire talented software engineers because they're so in demand. I would be glad to consider eager and talented prospective employees who I don't have to compete hard for.

I agree with this. If we're talking about the same thing, many of these are "Machine learning for hackers" type articles and only give a superficial exposure of machine learning with the implicit promise of quick and easy mastery. The problem is, there's really only one way to master this stuff and it's to open up a textbook and study hard. You really need to get a rigorous treatment of the theory to understand machine learning.

I'm not trying to be offensive, but what point are you trying to make? It sounds like you disagree with the author but your commentary seems like incoherent rambling to me.

I think the author makes excellent points.

It's hard to build a general AI business. E.g. a computer vision API provider. The technology is so democratized that you can't compete on algorithms alone. These APIs/services are more or less commodities nowadays.

Compare that to credit card fraud detection. For many years, there was one company/product (HNC/FICO/Falcon) that dominated the market (and largely still does) because they had a monopoly on the data. They smartly created a consortium and only they have the rights to train models on the data. They still use a relatively simple feedforward neural network with a ton of hand-tuned features. This is an example of vertical AI that created a wildly successful company.

The problem with plain old stock is the tax implications. Suppose you join my company when our last round valuation was $10M. Let's also say you negotiated 1% equity. If we granted you 1% equity, you would have a $100,000 tax liability. We just gave you something worth $100,000 so you have to pay taxes on it. You'd be crazy to be willing to pay taxes on risky equity that's theoretically worth $100,000. The company won't pay your taxes for you either. That's a huge waste of money. That's why stock options exist, to give employees the upside and minimize the tax liability.

Startups also like options because they believe it creates an "ownership mentality" among its employees. I believe this is mostly true.

I think equity compensation is also a selection mechanism. If I'm running an early stage startup, I want everyone to have a stake in the game. Equity compensation attracts employees with that mindset. Conversely, if a potential employee would prefer all cash compensation to equity, that would be a big red flag to me.

One of the struggles of offering equity to employees is finding a mechanism that has no taxable value upon issue, benefits from capital gains, and is legally sound.

One option is to organize as an LLC and offer a profits interest. These can be issued with $0 taxable value and benefit from the upside of the company. They can vest, and once vested they can participate in the gains of the company (including distributed income, not just a sale). I believe these are inherently more fair to the employees because there is no golden handcuff. They don't need to be exercised and once they're vested, you can walk away with them. On the downside, they are a little more cumbersome to set up.

Robert Naviaux is the real deal and his theory on the Cell Danger Response is fascinating [1]. In addition to ASD, he recently made a significant discovery in the characterization of Chronic Fatigue Syndrome as a metabolic illness [2].

[1] http://naviauxlab.ucsd.edu/wp-content/uploads/2016/09/cell_d... [2] http://naviauxlab.ucsd.edu/wp-content/uploads/2016/08/PNAS-2...

This is not meant to be a dig but just an observation for you to think about. Your last two comments are about how it isn't or shouldn't be possible to make money.

It's true that market opportunities are hard to find, but you'll never find one if all you see is the "yes, buts". You miss 100% of the shots you don't take.

For anyone who wants to learn linear algebra, I highly recommend Gilbert Strang's book and course from MIT: https://www.youtube.com/watch?v=ZK3O402wf1c.

I think what is needed is to understand what factors (e.g. environmental, genetic, etc) lead to autism so we as a society and parents can try to prevent it.

The amazing thing about this new wave of innovation in AI is that the technology is democratized -- anyone can have near state-of-the-art technology to power their applications without paying for it.

One of the lasting principles in the "business of AI" is that the data is the true source of value. If you don't have a monopoly on the data, you're going to have a tough time sustaining a competitive business because everyone else will more or less be able to clone your technology. My PhD advisor taught me this based on his experience founding one of the first really successful neural networks companies, HNC, which was later acquired by Fair Isaac (FICO). HNC put together a consortium of banks in order to train their credit card fraud detection system and maintained a monopoly on the data. FICO still runs this business today and I've heard from recent employees it still runs on the same old single-hidden-layer multi-layer perceptron architecture as it did in the 90s.

How many startups have you worked for that are turning a profit? Most technology ventures less than 5 years old are running in the red or they are reinvesting their profits to grow faster.

SportsPay (https://sportspayhq.com) | Software Engineer | Pasadena, CA | Full-time , On-site

We help youth sports organizations collect money efficiently from parents so coaches can coach and players can play. Our web app adds useful workflow on top of payment processing to save teams a ton of time and money.

We are funded, revenue positive and scaling our product and customer base.

We're looking for a software engineer with an ownership mentality who is willing to own the full stack from development to deployment and testing: Python/Flask, HTML/CSS/JS, PostgreSQL, Heroku, TravisCI, Github. Bonus points for candidates who have been technical co-founders or taken a lead technical role in early stage startups in the past.

Interested? [email protected]