- 14
- 19 213
AI Researcher
Portugal
เข้าร่วมเมื่อ 31 มี.ค. 2024
Hi all,
I am Manisha Sirsat, a senior researcher in Artificial Intelligence, holding a PhD in machine learning algorithms from University of Santiago de Compostela, Spain. With over 10+ years of hands-on experience in various areas of artificial intelligence, including machine learning algorithms, deep learning, computer vision, large language model, database management systems, etc.
The "AI Researcher" channel serves as a bridge between theoretical foundations and practical applications in the fields of machine learning, deep learning, large language models, computer vision, and many more. We are not just about understanding AI; we are about creating it, testing it, and pushing its boundaries. So, join me for a deep dive into AI research, hands-on implementations, and insights that light the path for future technologies.
Together, Let's explore the intelligence.
I am Manisha Sirsat, a senior researcher in Artificial Intelligence, holding a PhD in machine learning algorithms from University of Santiago de Compostela, Spain. With over 10+ years of hands-on experience in various areas of artificial intelligence, including machine learning algorithms, deep learning, computer vision, large language model, database management systems, etc.
The "AI Researcher" channel serves as a bridge between theoretical foundations and practical applications in the fields of machine learning, deep learning, large language models, computer vision, and many more. We are not just about understanding AI; we are about creating it, testing it, and pushing its boundaries. So, join me for a deep dive into AI research, hands-on implementations, and insights that light the path for future technologies.
Together, Let's explore the intelligence.
What is Prompt Design and Engineering? (Practical Implementation)
#promptengineering #promptdesign #llm #deeplearning #machinelearning #ai
In this video, I introduced prompt design and engineering and showed its implementation. We will explore fascinating concept of prompt design and engineering, a key component in the field of artificial intelligence. Prompt Engineering involves developing inputs that guide large language models or LLMs, to generate desired outputs. The quality and structure of these prompts significantly influence the efficiency and accuracy of the responses we obtain from LLM systems.
Github repository for the code: github.com/manishasirsat/prompt_llm_examples
--------------------------------------------------------------------------------------------------------------------------------------------------------------
Generative AI Playlist: th-cam.com/video/ID04YmgzM38/w-d-xo.html
Deep Learning Playlist: th-cam.com/play/PLzkBTicHqQFmY89SS1Xkfe-gpGCr-XGJo.html&jct=a2fQJbo-0p7KqCRdDNC9lSdLYCPcag
--------------------------------------------------------------------------------------------------------------------------------------------------------------
Connect with me on social media platforms:
Website: ai-researchstudies.com/
Google scholar: scholar.google.com/citations?user=kM4QN-8AAAAJ&hl=en
LinkedIn: www.linkedin.com/in/manishasirsat
GitHub:github.com/manishasirsat
Quora: machinelearningresearch.quora.com/
Blogger: manisha-sirsat.blogspot.com/
Twitter: ManishaSirsat
⏱️ Timestamps
0:00 Intro
0:35 What is prompt engineering?
1:38 What is prompt?
3:06 How a prompt works?
3:48 Why is prompt design important?
4:31 Key elements of prompt engineering?
5:30 Prompt examples
8:54 Practical implementation | prompt examples
In this video, I introduced prompt design and engineering and showed its implementation. We will explore fascinating concept of prompt design and engineering, a key component in the field of artificial intelligence. Prompt Engineering involves developing inputs that guide large language models or LLMs, to generate desired outputs. The quality and structure of these prompts significantly influence the efficiency and accuracy of the responses we obtain from LLM systems.
Github repository for the code: github.com/manishasirsat/prompt_llm_examples
--------------------------------------------------------------------------------------------------------------------------------------------------------------
Generative AI Playlist: th-cam.com/video/ID04YmgzM38/w-d-xo.html
Deep Learning Playlist: th-cam.com/play/PLzkBTicHqQFmY89SS1Xkfe-gpGCr-XGJo.html&jct=a2fQJbo-0p7KqCRdDNC9lSdLYCPcag
--------------------------------------------------------------------------------------------------------------------------------------------------------------
Connect with me on social media platforms:
Website: ai-researchstudies.com/
Google scholar: scholar.google.com/citations?user=kM4QN-8AAAAJ&hl=en
LinkedIn: www.linkedin.com/in/manishasirsat
GitHub:github.com/manishasirsat
Quora: machinelearningresearch.quora.com/
Blogger: manisha-sirsat.blogspot.com/
Twitter: ManishaSirsat
⏱️ Timestamps
0:00 Intro
0:35 What is prompt engineering?
1:38 What is prompt?
3:06 How a prompt works?
3:48 Why is prompt design important?
4:31 Key elements of prompt engineering?
5:30 Prompt examples
8:54 Practical implementation | prompt examples
มุมมอง: 224
วีดีโอ
Convolutional Kolmogorov-Arnold Networks: Introduction and Implementation
มุมมอง 592หลายเดือนก่อน
#convolutionalKAN #kolmogorov #kan #ckan #neuralnetworks #deeplearning #machinelearning #ai In this video, I introduced Convolutional Kolmogorov-Arnold Networks (Convolutional-KAN) and showed it’s implementation. Convolutional-KAN is an innovative alternative to the standard Convolutional Neural Networks (CNNs) that have revolutionized the field of computer vision. Convolutional Kolmogorov-Arno...
Convolutional Neural Network (CNN): Introduction and Implementation
มุมมอง 7453 หลายเดือนก่อน
#cnn #neuralnetworks #deeplearning #machinelearning #ai In this video, I introduced convolutional neural network (CNN) and showed it’s implementation. CNN is a type of deep learning algorithm mainly used for processing and analyzing image data by automatically learning spatial features through backpropagation. It consists of layers such as convolutional layers, pooling layers, and fully connect...
How to access GPT Large Language Model via OpenAI's API? (Practical Demo)
มุมมอง 4033 หลายเดือนก่อน
#openai #openapi #chatgpt #llm In this video, you will learn how to access GPT via OpenAI's API and see a practical demo on making API call. Github repository for the code: github.com/manishasirsat/access-llm-openAI Generative AI Playlist: th-cam.com/video/ID04YmgzM38/w-d-xo.html Deep Learning Playlist: th-cam.com/play/PLzkBTicHqQFmY89SS1Xkfe-gpGCr-XGJo.html&jct=a2fQJbo-0p7KqCRdDNC9lSdLYCPcag C...
Multilayer Perceptron (MLP) Neural Networks: Introduction and Implementation
มุมมอง 2.2K3 หลายเดือนก่อน
#mlp #neuralnetworks #deeplearning #machinelearning #ai In this video, I introduced the concept of multi-layer perceptrons and demonstrated its implementation. MLP is a specific type of neural network used for tasks such as pattern recognition and predictive analysis. Github repository for the code: github.com/manishasirsat/mlp Generative AI Playlist: th-cam.com/video/ID04YmgzM38/w-d-xo.html Co...
KAN Practical Implementation (Kolmogorov-Arnold Networks Algorithm)
มุมมอง 3.4K4 หลายเดือนก่อน
#kan #Kolmogorov-ArnoldNetworks #mlp #deeplearning #machinelearning #ai In this video, I tried to implement Kolmogorov-Arnold Networks (KAN) Algorithm using imodelsx library. The KAN is an approach in the field of machine learning that is based on the Kolmogorov-Arnold representation theorem from mathematical analysis. This method applies the theorem's insights to build predictive models for co...
KAN: Kolmogorov-Arnold Networks Paper Explained
มุมมอง 3.5K4 หลายเดือนก่อน
#kan #mlp #deeplearning #machinelearning #ai In this video, I explained the recent research study that is Kolmogorov-Arnold representation theorem. The KAN is an approach in the field of machine learning that is based on the Kolmogorov-Arnold representation theorem from mathematical analysis. This method applies the theorem's insights to build predictive models for complex, high-dimensional dat...
How to access LLMs from hugging face? (Practical Demo)
มุมมอง 3.2K4 หลายเดือนก่อน
#llm #languagemodel #huggingface #gpt Learn how to use Hugging Face and to access LLMs (Large Language Model) from Huggingface. Github link to access the code from the video: github.com/manishasirsat/access-llm-huggingface Generative AI Playlist: th-cam.com/video/ID04YmgzM38/w-d-xo.html Connect with me on social media platforms: Website: ai-researchstudies.com/ Google scholar: scholar.google.co...
How to Integrate RAG - Retrieval Augmented Generation into a LLM? (Practical Demo)
มุมมอง 1.9K4 หลายเดือนก่อน
#RAG #Retrieval-AugmentedGeneration #llm #largelanguagemodels #languagemodel #gpt In this video, you will see a practical demo on how Retrieval-Augmented Generation (RAG) can be integrated into a Large Language Model (LLM). Github repository to access the code: github.com/manishasirsat/rag-llm-demo What is Retrieval-Augmented Generation (RAG) Architecture: th-cam.com/video/fjNVID5Q9YA/w-d-xo.ht...
What is Retrieval-Augmented Generation (RAG) Architecture?
มุมมอง 6964 หลายเดือนก่อน
#RAG #Retrieval-AugmentedGeneration #llm #largelanguagemodels #languagemodel #gpt In this video, you will see the differences between a standard Large Language Model (LLM) and a Retrieval-Augmented Generation (RAG) model and architecture of RAG from the two research papers. 1. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks: arxiv.org/abs/2005.11401 2. Retrieval-Augmented Gener...
History of Large Language Models (LLMs) | From 1940 to 2023
มุมมอง 4264 หลายเดือนก่อน
#llm #largelanguagemodels #gpt #languagemodel Historical overview of the development of language models, from 1940 to 2023. Starting with the idea of Artificial Neural Networks (ANNs) introduced by Warren McCulloch and Walter Pitts in 1940, the timeline moves through significant advancements in the AI. This progress shows the rapid advancement in language models over the years. Thank you!:) Gen...
Understanding Transformer Architecture of LLM: Attention Is All You Need
มุมมอง 8314 หลายเดือนก่อน
#transformers #llm #architecture #gpt #largelanguagemodels The research paper "Attention Is All You Need" introduced transformers, which have transformed the way machines comprehend and produce human language. As a significant improvement on older models such as RNNs, LSTMs, and GRUs, transformers distinguish themselves by their ability to process data in parallel rather than sequentially. This...
The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
มุมมอง 9684 หลายเดือนก่อน
#1bit #llm #largelanguagemodels #nlp #gpt #microsoft The paper, The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits discusses a significant advancement in the field of large language models where these complex AI systems are being optimized to operate using only 1.58 bits. The 1.58-bit LLM defines a new scaling law and opens the door for new hardware and optimization algorithms. Y...
make a video on fine tuning ai
Thanks for the suggestion! A video on fine-tuning LLMs is definitely coming soon...
Good Video
Thanks
Maam kindly give detail lectures list which teach us how i can make apps by using Generative Ai and Hugging face or some free recourses if anyone cannot gain paid chatgpt API's
Sure, I will try to provide it asap...
@@airesearcher24 Thanks
Today i learned something new. Thankx Manisha
Glad to know that 👍
@@airesearcher24 Thanks for the great video.👏 But, the same thing i have done myself. when i am asking the very simple question the response from the gpt2 is so weird..... why it so?
Hi, you are right 👍 GPT-2 is an older LLM which can cause such issues. You can consider replacing GPT-2 with Llama2/Llama3/any updated LLM instead...
Thanks for the video. For the first feature x0,1 we have 5 features for the same input x0,1 how the output is going to be different although they used the same input, grid size, degree and knot vector?
I think, the five outputs (phi_{0,1,1}, phi_{0,2,1}, phi_{0,3,1}, phi_{0,4,1}, phi_{0,5,1}) for the input x_{0,1} are different because each neuron has its own distinct set of learnable/trainable coefficients... These coefficients adjust the B-spline activation functions in each neuron which allows each to respond differently to the same input based on what it learns during training...
Great video! How did you calculate the number of input features for the linear layer? Why is it 625?
I used the 'KAN Conv MLP2' architecture as outlined in the referenced paper. The authors developed this architecture through trial and error process. However, you have the flexibility to tweak this architecture's hyper-parameters to better fit your specific needs. For more detailed information, you can review the paper (arxiv.org/pdf/2406.131559).
व्हेरी गुड बेस्ट ऑफ लक मनीषा
Thank you very much for this great lecture video!! Save me a lot of time!!!
Glad it was helpful:)
Precise
This is lovely, actually I am doing similar MLP-replacing research and I wonder what dataset did you use for this experiment. Because I believe KAN seems promising but the reasons the paper insists whats better than MLP is actually not a problem of it for most cases.
It's great to hear you are working on KAN-MLP research. In the video, I used MNIST dataset. Here is some code that might be useful: github.com/manishasirsat/Convolutional-Kolmogorov-Arnold-Network/blob/main/convolutional-KAN.ipynb
@@airesearcher24 Thank you
Finally, Had been waiting for 2 months.
Great video!
Glad you enjoyed it
Artificial super intelligence vs artificial general intelligence
HThank you so much mam for valuable timeand information and asking one more favour , help me if you have time , i am trying publish a paper on human brain and image processing
Can you suggest some reserch papers which relate to the topic belongs how human brain process images
Can you suggest some papers related to topic " ai and human brain "
Might be an interesting paper on brain image processing: www.sciencedirect.com/science/article/pii/S1053811917308613
Amazing its like encyclopedia it helps me a lot thank you for your valuable time
Informative video, good luck and Keep it up
Thanks :)
No one can denay the importance of KAN as an efficient with more deep learning esp in multivariate time series analysis and forecasting with activation functions at nodes .A good and concept clearing presentation.
Yes, you are right
Good job.Very useful in my Telecom Research.Thanks.
Glad it was helpful!
Can you suggest a paper best introduction to deep learning ?
Here are couple of interesting papers: 1) link.springer.com/content/pdf/10.1007/s42979-021-00815-1.pdf; 2) hal.science/hal-04206682/document
1) why didn't you use the LBFGS method?
Mam, can you please tell me the list of free models & their size are small means compatible to run in google collab, used for prompting but I tried many & gets crashed like mistral , llama 2 due to all RAM used.
Great, you are interested in trying this. Here are some smaller models that you could try: DistilGPT-2; BERT Small; RoBERTa Base; T5 Small; BERT Mini; GPT-2 Small
waiting for RNN 😄
your explanations shows how clear your fundamentals are, I'm amazed....really enjoying watching all your videos 💯
thanks:)
Very good video, lots of love from Lahore Pakistan
really informative and concise :D
Thanks:)
explain with face cam so I can understand deeply, as the facial expressions contribute to the weight of neural network that is forming inside my brain and I want to make these concepts to be stored permanent in my head I prefer strong flashing images which can be stored in my neurons easily. Thanks
I am going to try that soon. Thanks for the input :)
Good explanation - I am trying to use in my Telecom Project......Thanks a lot.
Great!
heya we cannot get the answer if we haven't paid for the credits right?
Yes, that's correct. You need to pay for credits to use the OpenAI's models. For free access to some LLMs, you can refer to this video: th-cam.com/video/c0PXr1Eazr8/w-d-xo.htmlsi=qrpA4r7sAHbAeZjE
thanks yar
To solve the imodelsx import error, please try these commands: 1. %pip uninstall imodelsx 2. %pip install imodelsx==1.0.5 --force-reinstall
Thank you for your great Video! Unfortunately the commands doesn't solve the import error for me :( Do you have another idea?
Ok, I am going to check it again..
thanks a lot this is really helpful
Glad it helped.. 👍
Thankyou vahini keep it up👏👏
Thanks:)
good explanations then our professors in our colleges wish you further bright success
Thanks you:)
Thanks a lot for this informative video... The first Convolution example would have output at [0,0] as 7 instead of 8... Please recheck...
Yes, you are right, Thanks for noticing it 👍
Can you please tell me whats the motto of this channel? Im pursuing ECE. Currently in my 2nd sem. And am really into AI/ML. If you have some spare time could you please tell how can I make best use of these informative videos? I would really love to know how would you advice me to attend the videos.
Hi, the motto of this channel is to connect theoretical foundations with practical applications in AI, covering machine learning, deep learning, large language models, computer vision, and more.. To make the best use of the videos, take detailed notes, practice coding exercises, and work on small projects to apply your knowledge... Stay consistent for the best learning experience:)
👍
Really Awesome explanation. Thanks for your time and effort. You are simply the best about providing deep learning concepts.
Thank you:)
This may be just the best treatment of the subject on TH-cam. Thank you. Well done.
Thank you! Glad you found the video helpful.. Your feedback means a lot to me:)
I am facing an ImportError in the line: "from imodelsx import KANClassifier" Any suggestions here?
yeah me too i have the issue as well
Let me check the error and I will get back to you..
@@airesearcher24 Thanks for the reply. The exact error that is showing is as follows - ImportError: cannot import name 'indices_to_mask' from 'sklearn.utils._mask' (/usr/local/lib/python3.10/dist-packages/sklearn/utils/_mask.py)
@@airesearcher24 Hi, I hope you're doing well. I'm currently stuck on a portion of the tutorial due to the error and was wondering if the issue has been resolved. Any guidance would be greatly appreciated!
Guys, please try these commands. Hope this will solve the error:) 1. %pip uninstall imodelsx 2. %pip install imodelsx==1.0.5 --force-reinstall
Very nice my dear friend Please give me your number I need you in a question
Hey manisha, do check your mail. I have mailed you regarding having a podcast with you.
Hi, I do check. If I missed yours, I will take another look..
Good one.
Thanks!
I have a project code, wanted to make an AI to index my project files and analyse the code in local environment. On final which can give response based on prompts. Can you make a video on that?
Sure, I will make a video on that. Thanks for the suggestion. Keep watching:)
Fantastic ! My background is Math, not coding so this was a great way for me to understand the 'Attention is all you need' paper.
Glad you enjoyed it! Stay tuned for forthcoming videos:)
Great video. This was a good summary of the progression of LLMs.
Thank you:) keep watching 👍
Hi. Manisha. Great content. Hope we can chat in the email as we are searching for an AI researcher that can built models that can be applied to financial markets. Thanks and warm regards. Lim
Hi Lim, thanks for reaching out. Your interest is appreciated.
Love you mam ❤
I'm binge watching to catch up on your content.
Great, enjoy catching up on the content and feel free to reach out if you have any questions/comments.
Great channel! These are some of the *best* videos on the topic! Subbed!!!
Glad that you are enjoying the content. Welcome to the community:)
Given the propensity for gradient vanishing in deep recurrent neural architectures, how would one hypothetically mitigate catastrophic forgetting in a sequence-to-sequence model while maintaining stochastic gradient descent convergence and ensuring orthogonality in the weight matrices of a bidirectional LSTM with layer normalization?
Hi, I think to address catastrophic forgetting in a bidirectional LSTM with layer normalization while ensuring SGD convergence and maintaining weight matrix orthogonality, implement Elastic Weight Consolidation for stability, use orthogonal regularization, adjust learning rates, and apply gradient clipping. Hope this paper might be helpful: Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., ... & Hadsell, R. (2017). Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114(13), 3521-3526.
@@airesearcher24 Thank you
it's deep understanding of transformers clear way !!
Glad it helped:) keep watching 👍