@EverLearner2

NLP Researcher, NLP Evangalist, Thought Leader https://xitter.cf/t.co/Jr4813FkL0

Joined July 2020
  • Tweets79
  • Following123
  • Followers19
10Media
Humans are excellent at few-shot learning. Giant GPT-3 showed hope. Can models be greener & better? Pattern exploiting training (PEP) trains a few smaller models with few shots, yields a soft labeled dataset with distillation, beats GPT-3. @timo_schick @HinrichSchuetze #NLP
1
1
The paper for reference arxiv.org/pdf/2009.07118.pdf
1
Nudge & budge, that's how soft prompting stimulates brain activity. The new line of research where language models have gone from uni-task capability to multi-tasks with "prefixs", "prompts" or "soft prompts". soft prompts awaken otherwise frozen models. #NLProc #DeepLearning #AI
1
1
@colinraffel, @blester125, @victorsanh @huggingface - happy to see intuitive but profound work of yours carving this much-needed space.
Ability to generalize stems from the ability to multitask. It's relatable to how a baby learns. Looks like a small step towards Artificial General Intelligience. Check out its empirical evidence -arxiv.org/pdf/2110.08207.pdf #ArtificialIntelligence #Deeplearning #NLProc #DataScience
1
Lets take a shot at a famous question, Is Logistic Regression a Classification or a Regression technique? It is classification in the space of input, but regression in the space of output probabilities. What do you think? #MachineLearning #DataScience #MLfundamentals
1
Once achieved, it looks so obvious, and joy fades away! The real joy experienced is during the journey because of inherent uncertainty! Keep moving! #phdlife In hindsight, Relu looks so simple but wasn't the case when ANN was tired of trying sigmoid for a decade. #DeepLearning
1
1
The future of sci-fi movies lies in #Facebook metaverse powered experience. Each of us would assume a character while we what a movie. Imagine the star trek or titanic experience. #AI
1
The world appears non-parametric! Smart mathematicians reduce any such world to a parametric distribution so that it gets simple and well behaved. So is our life, did we reduce it to a set of parameters? Do we call them values? What are they? #LifeLessons #Maths #DataScience
1
#NeurIPS2020 and #ExplainableAI, Empirical validation of DNN explanation methods suggests 1) Explanation-by-example is the most preferred method for images, audio & sensory data. 2) LIME (i.e. input perturbations) is a winner for explaining text-domain. tinyurl.com/y23uwc54
2
Data is never biased, it's always the sampling method that we choose to play with the data. #JustSaying #DataScience #artificalintelligence #Explainability #AIEthics
1
It's not about avoiding, rejecting, or ignoring something unimportant but it's all about chasing, indulging, and enjoying something important. #tuesdaymotivations @PhDForum @AcademicChatter
7 steps for #DataScience 1.Simplify & Formulate a problem 2.Define sub-modules & interfaces 3.Make/break assumptions 4.Rapid prototyping 5.Keep hypothesizing & experimenting 6.Analyse, infer & reason 7.Embrace for real-world application, if solution doesn't generalize, goto 5
Think of AI would learn to code algorithms, by merging computational ability of Von Neumann architecture and probabilistic inference of Deep Neural Networks. AI giants are in making. Baby step. @AcademicChatter #DeepLearning #ArtificialIntelligence #NLProc aclweb.org/anthology/D19-160…
Apple now has an AI/ML residency program! I'm looking forward to working with our first class of residents.
Michael Rennaker@MrRennaker27 Aug 2020
Thrilled to announce a new program designed to help experts in applied fields build ML-powered products and experiences. Introducing the AI/ML residency program: machinelearning.apple.com/up…
11
156
6
1,165
Do NLP models learn "meaning"? Can Octopus intercepting 2 conversing humans ever learn 'intent' beyond 'mimicking form' from hindsight data? Can form express true meaning? Nicely covered @ "Climbing towards NLU". bit.ly/32ii0vH #AcademicChatter #NLProc #DeepLearning
1
2
2
Never thought, muppets of Sesame Street would play a decisive role advancing #NLProc. ELMO, BERT, ERNIE, Grover, etc What next? I can suggest two, Rosita: RObust and Self Inductive Training Augmented Model, Zoe: Zen Optimised Embeddings Model #AcademicChatter
#ShameOnMumbaiPolice
Subramanian Swamy
@Swamy3930 Jul 2020
Why I think Sushanth Singh Rajput was murdered
Math analogies are insightful. A highest rank matrix has a unique solution, so the highest rank people (achievers in their own fields) tend to have a (arguably) unique solution towards life, and remain far less distracted (empirically). #Mathematics #AcademicChatter #LifeLessons
1
New to #ArtificialIntelligence and #NeuralNetworks and #DeepLearning, this is an amazing illustrative resource to pick up the pace neuralnetworksanddeeplearnin…
1
1
2
Load more