Sunday Reads – Links I Found Interesting 6/16

PyTorch

PyTorch Hub is an awesome new step forward in research reproducibility for machine learning and artificial intelligence. Makes it super easy to publish your pre-trained research models and for others to download them and test them. I was shocked how uncommon it was for people to publish their models when I first started reading machine learning papers.

An enzymatic pathway in the human gut microbiome that converts A to universal O type blood

This was a metagenomic study that identified two enzymes that may help convert A and B blood to O increasing the supply of blood available.

SciHive

This is a cool ArXiv access point with upvoting and down voting, ability to save notes, see twitter commentary on paper and more.

Advertisements

Sunday Reads – Links I Found Interesting 6/9

Mapping human microbiome drug metabolism by gut bacteria and their genes

A fascinating look at how the microbiome may affect drug metabolism. Important to remember that the game does not end at pharmacogenomics and we need to be paying attention to the complex interplay of numerous complex systems to understand drug action.

Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer

Thanks to the ‘magic’ of deep learning we may be able to better predict which patients are going to respond to immunotherapy in gastrointestinal cancer with cheaper tests. More people treating their cancer certainly sounds good to me.

CCR5-∆32 is deleterious in the homozygous state in humans

The gene that was CRISPR-ed in those Chinese babies makes it more likely you die. This isn’t even accounting for the potential off-target effects. Turns out the thing we all knew was unethical is in fact unethical. Who da thunk?

Principles of and strategies for germline gene therapy

Following in the same vein as the previous article we take a theoretical look at the potential for these germline therapies.

The Sweetgreenification of Society

Interesting Substack post about the increasing stratification of society through the lens of boutique businesses.

RNA sequence analysis reveals macroscopic somatic clonal expansion across normal tissues

From one, many. Our bodies are a huge mess of different mutations each of which could or could not be maybe contributing to diseases. Thinking of yourself as having one genetic identity is flawed.

A Jaunt Down Financial Fraud Lane

A fun article taking a look at some of the numerous scams in the cryptocurrency ecosystem. I am partial to the disaster that is EOS.

Subscribe at https://bennettftomlin.substack.com/

Sunday Reads – Links I Found Interesting 6/2

EfficientNet: Improving Accuracy and Efficiency through AutoML and Model Scaling

Google did that thing they do again where they make vast steps in artificial intelligence and machine learning. Efficiency of these image recognition networks is up to 10* greater. Most of this gain is because they try to use “scaling coefficients” so that the network scales in a predictable way. I’m just mad because it’s a TensorFlow model and not PyTorch so I can’t drop it into any of my existing image recognition notebooks.

A promising step forward for predicting lung cancer

Another Google blog post about how they are doing incredible things. Man what I wouldn’t do to work for Google Brain. (This research is also being done at the same university I am doing my capstone with, so hey maybe they can sneak me in) Okay so in this article they describe a state of the art result for predicting lung cancer using improved volumetric predictions of CT scans. They instead of looking at individual slices in the image are instead reconstructing 3-d structures to improve the accuracy. This both is and is not a crazy step forward. Being able to use the 3-d structure seems to be truly revolutionary, but some of the radiologists performed equally as well as it. Seems that it will be a useful assistance tool for now.

Moving Camera, Moving People: A Deep Learning Approach to Depth Prediction

I promise this won’t be all Google, but again what they are doing right here is incredibly cool and a little bit scary. They have found a way to approximate the 3-d size, shape, and depth of moving people even when the camera is moving. This work has really cool implications for AR and VR and a little bit terrifying uses for a potential police state. There are many places where face recognition has been banned or people are considering banning it, however, combining a 3-d map of a person with existing effective identification techniques like gait tracking can serve as a proxy for facial recognition in those areas. Combined with facial recognition it could provide an even stronger match limiting false positives, and avoiding false negatives.

Speech2Face: Learning the Face Behind a Voice

Okay we are finally away from Google, but into something even more terrifying. This neural network when fed a small sample of speech is able to generate a qualitatively accurate facial guess. The model seems quite adept at identifying both race and gender. Scary stuff.

Defund Crypto

This fun parody site created by Joshua Davis, Kyle Gibson, and the pseudonymous Cas Piancey mercilessly lampoons the tomfoolery of Kik’s attempt to challenge the SEC. For the record, I do not think promising an Ethereum public DApp and delivering a one node Stellar fork is a good thing.

Exist 

This interesting service will likely not be appreciated by the privacy minded. While they do have a strong privacy policy its purpose is to bring a ton of your disparate personal data together and find interesting correlations. Whether it is useful or just noise remains to be seen.

If you enjoyed this please subscribe to my newsletter at https://bennettftomlin.substack.com/

Expertise, Diminishing Returns, and Intersecting Niches

I have been thinking recently about skill development, because I have a habit of never being focused in a single area and instead trying to learn as much as I can in as many niches as I can. Because of this, I am only what could be considered an expert in a relatively narrow subset. However, I am significantly knowledgeable in a wider array of niches. I am beginning to think that this form of knowledge may actually be more valuable. (Crazy that I would try to justify my own shortcomings right?)

I also want to try to treat these mathematically in order to explain my thinking, however, do not worry the math will be kept relatively simple.

So this idea came to me after I was sitting and reflecting on the kinds of tasks that will become valuable as our economy progresses with increasing automation. The conclusion I came to is that many jobs now will be either partially or completely automated. Many industries where there is little fear of automation, will end up being partially automated, destroying job prospects and wages. For example: accounting. It likely will not be fully automated anytime soon, however, a significant number of hours are spent on heuristic based tasks that computers will be able to in large part supplant. There will still need to be humans, but if one person can do the work of ten wages will fall and job prospects will dissapear.

So what will become valuable? Until there is significant progress on Artificial General Intelligence, a huge amount of the automation will be in very focused niches. So the value will come from humans who can bridge niches, understand broader pictures and connect information from disparate realms. Now the issue with this at first glance seems to be that it is going to require significantly more time and effort to reach a competitive level in multiple niches rather than just one. I believe that may be a little bit simplistic however, and I will explain why.

There is this concept when you are learning a new skill called the point of diminishing returns. Basically as we asymptotically approach expertise the amount of effort required to gain additional expertise is exponentially greater. Or to phrase it more simply the vast majority of the improvement comes from the initial investments of efforts. The move from middle 50% to 95% may take the same amount of effort as the move from 95% to 99% which will take the same effort as the move to 99.9% which will take the same effort as the move to 99.99%. Why is this valuable to us? Because by focusing on intersecting niches we do not the same super high level of expertise in order to be successful.

Let me explain: expertise in intersecting niches can be expressed as the product of your expertise in each individual niche. So let’s say for example I am in the top 5% of the world in knowledge about healthcare, and in the top 5% of the world in knowledge about running a non profit, for expertise in running a healthcare based non profit I am not in the top 5% I am instead in the top 0.25%. How is this possible? Because our expertise in this intersection is equal to the product so in this case: total expertise=.05 * .05 = .0025. If I am correct about this construction of expertise (and I am not, it is vastly simplified but focus on the concept) then if we can bring even more niches into our areas of expertise we reach progressively rarefied realms of cumulative expertise.

To really emphasize this, consider the fact that we have already established that it will take the same amount of effort to reach the top 1% of a single niche, as it will take to reach the top 0.25% of the intersection of two niches. If I am correct about this vision of the future the value of the Renaissance man is back. Being able to conceptualize and view the world through multiple well honed viewpoints and see the connections between them will become an incredibly valuable skill.

If I am wrong, I have a whole list of skills where I am almost good enough to be an expert but not quite.