Flatiron Application
If there’s one thing I’ve learned from my internships, it’s this—I’m most productive and engaged with my work when my own morals and values are completely in sync with the company’s culture. Everything else follows from those first few well-defined principles.
I feel that way about Flatiron’s values—as someone whose two current biggest goals in life are to improve every day and be a good person, I see how I would be totally at home with “Be kind” and “Learn, teach and grow” as guiding directives at my workplace. I’m especially enamored by “Seek feedback at 30%” as a core value—in my prior internships, I’ve sometimes failed to seek feedback early, being hesitant to push out unfinished work, and only realized later how much smoother things would’ve been if I had iterated quicker. I’m really impressed that Flatiron has identified the importance of that and made it a company-wide cultural mandate, so no one feels afraid or awkward to ask for feedback when their work isn’t quite finished. To me, that’s the sign of a company that tries as hard as it can to create a productive and fulfilling workplace for its employees, which is why I’m really excited about Flatiron.
uid: 202009211522 tags: #applications
Actor-Critic
Two main components in policy gradient are the policy model and the value function. It makes a lot of sense to learn the value function in addition to the policy, since knowing the value function can assist the policy update, such as by reducing gradient variance in vanilla policy gradients, and that is exactly what the Actor-Critic method does.
Actor-critic methods consist of two models, which may optionally share parameters:
Critic updates the value function parameters w and depending on the algorithm it could be action-value Qw(a|s) or state-value Vw(s). Actor updates the policy parameters θ for πθ(a|s), in the direction suggested by the critic. Let’s see how it works in a simple action-value actor-critic algorithm.
Initialize s,θ,w at random; sample a∼πθ(a|s). For t=1…T: Sample reward rt∼R(s,a) and next state s′∼P(s′|s,a); Then sample the next action a′∼πθ(a′|s′); Update the policy parameters: θ←θ+αθQw(s,a)∇θlnπθ(a|s); Compute the correction (TD error) for action-value at time t: δt=rt+γQw(s′,a′)−Qw(s,a) and use it to update the parameters of action-value function: w←w+αwδt∇wQw(s,a) Update a←a′ and s←s′. Two learning rates, αθ and αw, are predefined for policy and value function parameter updates respectively.
uid: 202009181412 tags: #knowledge
Tuberculosis
- I didn’t know that tuberculosis can produce fusion of vertebrae and deformation of the spine - it has a much wider impact than I previously thought.
- Pretty crazy that you can see evidence for TB in Egyptian mummies, in two ways - first, that the disease has been around for that long, and second, that the combination of archaeology and biology can detect its presence.
- Similar to a lot of the diseases that we’ve seen in this class, the rise of the present wave of TB can be attributed in part due to zoonotic influences (“town “dairies”). Seems like that’s universally one of the major reasons why a large outbreak/pandemic happens.
- Wow, it’s pretty disturbing how TB was fetishized in the 19th century. I wonder if there’s any other comparisons to modern-day diseases. Probably not to the same extent, because we have a much better understanding of medicine now than we did then.
- TB being regarded as a “Jewish disease” in the urban centers of New York and Boston is eerily reminiscent of how COVID-19 has been called the “Chinese plague” by President Trump. I wonder why this casual racism often seems to go hand-in-hand with major pandemics.
- It’s absolutely crazy that a TB infection could occur if as few as 5 bacteria reach the alveoli of the lung. 5 bacteria! No wonder a quarter of the world’s population is said to have the disease.
- It’s interesting how the human body’s immune response (macrophages) are the ones that in some sense making the infection much worse, by transporing the bacilli to other parts of the body by the lymph channels.
- TB can literally liquefy your lungs - yikes.
uid: 202009172206 tags: #mcb55
Japjot Happy Birthday Video
Yo, happy birthday bro. I’m really bummed that we haven’t seen each other as much this year, with me being abroad and then COVID, but in the long term this is just a blip in the horizon for us. From dicking around in comp civics and taking the Bart every day to Math 54 to now and our entire journey in between, the last 3 years rooming with you have been absolute gold. There’s no one else I would rather have lived with, no one else I would rather have made those memories with, and I’m sure this is just the start for a lifetime of moments like those. You know, I’ve seen firsthand just how much you’ve grown, and matured, and discovered yourself over the last 3 years, and I couldn’t be more proud of you, and proud to be your friend. Here’s to hoping we’ll live again this semester, and if not this semester then definitely next semester. So yeah, love you bro, always happy birthday again, and poop!
uid: 202009042038 tags: #relationships
[LEC] CS 375 (Preparing for successful discussion)
Prev meeting: [LEC] CS 375 202008281000
- They’ve already had lecture! Some presentation OK, but keep it informal
- Sends message that you want them to sit there and passively absorb
- Want to practice being a lecturer? Just ask!
- Speak clearly & brightly
- Ensure you are heard
- Project your voice (“Hit the back wall”)
- Record yourself (or have buddy record you from back of room)
Motivation => Learning
- More motivated students will spend more time and effort to learn hard stuff
- Social interactions help learning & motivation
- (Try activities for pairs, teams)
The re-vote after discussion with peers is pretty important! Transfer question: After you see a problem that you just saw, there’s another problem that’s not based off the same template, but it is somewhat related, this has even slightly better results than after discussion question
What to ask about, or how students learn
Students add knowledge to existing framework (constructivism)
They stumble if that framework includes inaccurate/incomplete assumptions
Review/ensure background/prerequisite info
Dividing material into discrete chunks improves student acquisition of material
Do students “know what they don’t know”?
Learners need feedback (more frequent is better) and review (more practice is better)
A few scenarios/suggestions
- Make the unit participation fine-grained
- Worksheets, good or bad?
- Try to cover too much => speak too fast
- (too much participation => finish late)
- Grad GSI vs Undergrad GSI: How collegial to be
uid: 202009041000 tags: #meetings #teaching
Torch And The Firehose Notes
The sandwich method for general principles
To present an abstract idea—a general formula, a general law, a theorem—a good way is to present in order: • An easy example illustrating the principle • The general statement and explanation of the principle • A harder example using the principle
The bird’s-eye view and review
Often students can’t follow an explanation because they can’t tell where it is headed, what the ultimate goal is.You know, but they don’t. So begin a complicated explanation by giving an overview of what it is you’re going to do—the purpose, the general method, how long it will take, whatever. Some teachers like to write this outline on the board. When it’s all over, turn back and survey the ground you’ve covered together, and congratulate them on having made it.
uid: 202008301309 tags: #teaching