Thank you for your efforts. I came across your paper/code and posted it here. I was looking to find a technique to cost optimise transformer based question and answering. Presently I am using CPU and getting a GPU is too costly on AWS.
Since I use high level code I don't understand the maths completely. However, I was wondering if your techniques can be beneficial on CPUs?
If I were to use this to improve transformer based architecture what should be my approach?
It should be possible to get large speedups on CPUs, but the trick will be gradually approximating each of the layers in the model (see my reply to sibling comment). It's not conceptually difficult, but will require a fair amount of C++ work to port the code to GPUs* for training; and it will probably go slower than dense ops on modern GPUs due to tensor cores not supporting our memory layout.
I think of this paper as the first in a two-part series, where the next one takes these fast ops and gets them working in full neural nets. (If anyone wants to do this project, happy to coadvise you / talk about it whenever; I won't have bandwidth to do it myself for the foreseeable future).
I have been through this and these are things that helped me improve.
1. Show up
Make sure you show up instead of avoiding people and events. Initially it can be difficult. You may be worried if you would mess up the conversation. You might be anxious you would run out of words to talk to. And it might turn to be a little embarrassing, that's alright. Doing more and more of those interactions kills your inner fear. Getting past your fear is the first step.
2. Observe & Listen
Observe how other people talk. If you like the way someone approaches others, copy it. Imitate and see how it works out. Be observant on how others feel while you talk. Are they bored? Are they eager to talk more? Are they trying to move conversation to some other topic? Usually shy people are within their head too much they don't notice the environment and react to it.
3. Find something common
I connect with a person by finding something in common between me and that person. It could be politics, religion, engineering, events, etc etc. Find things you both care about, conversation will flow naturally.
4. Give Compliments
If you like something about a person, give them compliment. Tell them details of what you loved the most. Be specific.
It would take time, through multiple iterations, you will improve :)
Best wishes, I am confident you will be successful.
Dead on. I was going to say something along the lines of "fake it till you make it." If you pretend to be socially comfortable long enough, you'll eventually convince yourself. But this comment lays out specifically how to do that.
i would add, "5. Ask questions." if you struggle to make conversation, find one thing about the person and ask them about it. how did they decide to do what they do for a living? what about that hat made them buy it?
Most people only see the AI/ML aspects or the ability to respond like human. But what's adding value is the chatbot's interface. Working for a chatbot company that hit $1M revenue I can say the chatbot interface solves a good problem.
There's a good chunk of people who are stressed in having to be comfortable with vide variety of website interfaces. For my dad who's over 60 years struggles with the government's interface to avail the pension scheme.
Chatbot or conversational interfaces have a repeating pattern. People are used to conversational interfaces because of SMS, iMessage, WhatsApp, FB, etc etc. Providing an experience through a chatbot interface reduces the learning curve for the end-user. You are presenting tiny chunk of information and even collecting one thing at a time. It's drastically less overwhelming.
It's not just for 60 years, I have friends who are around my age late 20s who finds it difficult to navigate through websites that have many offerings (Banks, Govt Services, etc)
A fellow orthodox christian here. I have been trying to cultivate the habit of Jesus prayer. But the journey is a struggle.
However, even in the fallen attempts to walk the orthodox way of life I occasionally sense the joy and warmth described in the spiritual texts about Jesus Prayer. Thus motivating me to continue the path.
At first I used an app on my phone to send myself random reminders to say the Jesus Prayer throughout the day. That really helped spur the habit along. Now, I am almost always saying it while I'm driving, or if I'm doing chores around the house, or if I'm walking around somewhere. At first, it felt a little odd, but now I find it much more rewarding than just letting my brain wander around in a forest of nonsense and ego-focused self-reflection.
They should have been nice, instead of communicating agree with their terms or go and delete your account.
It's a lesson. No matter how big you are.
1. Communicate for understandability.
2. Don't take your customers for granted.
3. Change is hard, nudge them towards it instead of a push.
I mentioned somewhere else here the libraries I used, search for jbullet with your browser and you'll find it. The rest I built myself. Clojurescript is just for the website. I don't have a background in mechanics, so no, not really. If anything I was trying to avoid the complications that real mechanical engineers have to deal with.
Since I use high level code I don't understand the maths completely. However, I was wondering if your techniques can be beneficial on CPUs?
If I were to use this to improve transformer based architecture what should be my approach?