Vishal Parashar

I talk to machines, of course using natural language.
Data Scientist | Tech Enthusiast

[Vishal Parashar @imVParashar]

Deep learning models really shine with large amounts of data.

Having enough labeled data is a constant challenge in the field, especially in NLP.

A successful approach that has yielded great results in the last couple of years is of 𝙩𝙧𝙖𝙣𝙨𝙛𝙚𝙧 𝙡𝙚𝙖𝙧𝙣𝙞𝙣𝙜.

Thread:👇

[Vishal Parashar @imVParashar]

@ramsri_goutham Congratulations! 🎉

[Vishal Parashar @imVParashar]

I just published Generating Important Video Clips from Lengthy Videos with AI https://t.co/1fOFPtpooM

[Vishal Parashar @imVParashar]

@ramsri_goutham Sounds interesting! I recently implemented it. I needed a json output from LLMs.
The struggle was that the output was not consistent. After a lot of experimentation with the prompt, made it to work. But I'm looking for a more neat way to do this.

[Vishal Parashar @imVParashar]

@ramsri_goutham @yenceesanjeev @nwbotha Congratulations Ramsri! More power to you. :)

[Vishal Parashar @imVParashar]

@capeandcode Certainly yes!

[Vishal Parashar @imVParashar]

@capeandcode Good Luck! :)

[Prashant @capeandcode]

MLOps and Challenges

• MLOps is gradually becoming a critical component in data science projects.

• For sustainable ML deployments, it is essential to understand and apply MLOps concepts.

Let's dive ↓

[Vishal Parashar @imVParashar]

I just published Implementing a Custom Intent Classification (Transformers DistilBERT) in RASA 3.X https://t.co/8ejZV2VUp5

[Vishal Parashar @imVParashar]

@punjabiii_munda Mark Zuckerberg to the engineering team.
#WhatsAppDown

[Vishal Parashar @imVParashar]

Mark Zuckerberg right now.
#WhatsApp #WhatsAppDown

[Tushar Raj Verma @TrajVoid]

When you want to present your Machine Learning Solution in various hackathons or POCs in significantly less time, with fewer lines of code and nice looking UI.

Here is a better idea instead of setting your solution in a notebook, Streamlit helps in a pretty good way with python.

[Tushar Raj Verma @TrajVoid]

What is a dead neuron?

Today’s thread is based on concepts, drawbacks, and rectification around the topic of the dead neuron.

Keywords of thread - Sigmoid, Vanishing Gradient, Relu, Leaky Relu, and much more

1/10

[Tushar Raj Verma @TrajVoid]

‣ AUC-ROC curves are crucial in deciding the threshold part for binary classification problems.

‣ In most cases, we take a 0.5 score as the threshold

‣ But what if I told you that there is a mechanism that is used to select the best threshold

Concepts and Visualization🧵

[Vishal Parashar @imVParashar]

Now, this is definitely the next level.

[Google AI @GoogleAI]

Introducing Imagen, a new text-to-image synthesis model that can generate high-fidelity, photorealistic images from a deep level of language understanding. Learn more and and check out some examples of #imagen at https://t.co/RhD6siY6BY

[Prashant @capeandcode]

Machine Learning | Mathematical Resources

Here are some of the resources that I went through for understanding mathematics relevant to machine learning understanding ↓

[Vishal Parashar @imVParashar]

@ramsri_goutham @ProductHunt @yenceesanjeev @nwbotha Congratulations! 🔥
Here's hoping this product gains a lot of attention :)

[Vishal Parashar @imVParashar]

With its viable framework for pre-training and transfer learning in NLP, ULMFiT was able to provide the missing element to take transformers to the next level.

[Vishal Parashar @imVParashar]

Universal Language Model Fine-tuning 📈

ULMFiT is a method to enable transfer learning for any NLP task, providing great results.

And all of this without having to train your models from scratch. Isn't it interesting?

Let's see how: 👇

[Vishal Parashar @imVParashar]

3. Fine-tune (fine-tuning the classifier on the target task)

In this step, the language model is fine-tuned with a classification layer for the target task (e.g., classifying the product reviews given by the users).