Interesting Niches - Simply Interests (Notes from March 30 to April 5, 2020)
Nat Eliason, of Roam and productivity-type videos fame, mentioned that consistency has been the primary driver for his content rise. It's refreshing to come across his newsletters as well as Tiago Forte and David Perrell's for how varied their content is. Typically, we don't have one specific idea we follow. It's a collection of all of our readings and experiences. My collection has probably been more evident in this - especially if you pay attention to the notes - generally around startups but also finance, vc, sports, media. Plenty that interest me.
So, this is hopefully my last Wordpress post on here and moving to Webflow to move my thoughts and musings for the recent time further.
In light of this, I am going to be doing something with lists. Specifically, around startups. But I've been obsessed with historical rankings / lists for a long while. Only now going to be putting the together (I lied - I've been piecing the together for a while). Hopefully it leads to explorations and transparency with connecting various lists around them. Be it industry, job hunt, maybe put some clarity to the lists that are typically without much detail. We'll see! Look out for further info.
![](https://cdn.prod.website-files.com/5fd1587a7d2fa2028f047746/60528c0355c742a33a77e5c1_Iguazio-logo-for-LinkedIn.png)
- Corona Investing (Meb Faber Podcast Part II, III)
- Staying rich and dealing with downtrends
- Cash / yield / gold all have drawdowns > 48%, so if you lose half of your money, does it matter?
- Mixture of trend following and yield can get drawdown to ~30%s
- 16min on the News
- Yaron Haviv, CTO of Iguazio & Mahesh (ML Ops Webinar 3/31/20)
- Develop and test locally and then turning it into production
- Package (dependencies, parameters, run scripts, build)
- Scale-out (load-balance, data partitions, model distribution, AutoML)
- Tune (parallelism, GPU support, Query tuning, caching)
- Instrument (monitoring, logging, versioning, security)
- Automate
- Streamlining collection of data, prepare at scale, accelerate training and deployment
- Why are ML Projects not deployed seamlessly? (Survey results)
- No starting with clear business obj - why are we doing this? Similarly, not a good business case
- Management failure including insufficient investment
- Poor communication or not having the right skills for the job
- Management resistance ("gut" and "real-world insight" over analytics and data)
- Selecting the wrong uses, especially in an overly ambitious project
- Data scientists asking the wrong questions due to lack of domain knowledge, primarily
- Disagree on enterprise strategy
- Big data silos
- Analytic Lifecycle - ML Eye
- Define business mission (eg - Reduce churn rate in cc usage by 15%)
- Project definition and resource evaluation (eg - Estimate propensity for cardholders to churn)
- Analytic solution design - translating objectives into data science tasks, workflow (eg design churn prediction solution)
- Capture and data preparation leading in to Algorithm Prototyping (eg prototype)
- ML Ops - key drivers for success for ML Platform
- Resource management (ability for multiple people to use multiple GPUs/machines running environment)
- Experiment management (ability to trace code, CL parameters, dataset for trained model - ability to keep track of result with envs)
- Capability to store models automatically, hyperparameter optimization (framework that helps search over optimal hyper param settings)
- Provision to store and manage ML datasets, models using tagging, automated versioning and querying capabilities
- ML Flows (drag and drop, visual tool to build pipelines), rapid experimentation, share & re-use
- Deployment
- Data science needs to quickly adapt
- What worked before won't work now or in the future - concept drift
- Need for fast, iterative changes
- Synthetic data to create a basis for models in times of uncertainty - no time to deal with complexities in deployment
- Need to see business impact quickly
- Wade Arnold, Founder of Moov.io with Sam Maule MP of North America at 11:FS (11:FS FinTech 4/2/20 morning)
- Services consuming vs what we're paying for
- Moov.io as free connection of services for banks using GitHub and pulling data from gov sites, for instance
- Project in Australia to update them
- For the big 3 core banking - they weren't to be used on the internet originally
- He helped add an abstraction layer
- He wants fintech to be more of a d2c term than services and layers in banking
- In the past, using Microsoft SQL or Oracle - now, you wouldn't want that
- All Open Source, especially internet and cloud providers enabling them
- Not much different than IBM adding mainframes before
- Open Source vs proprietary tech is scale
![](https://cdn.prod.website-files.com/5fd1587a7d2fa2028f047746/60528c04d2cf1c16aed52ce8_moov-share-image.png)