Dr. Deb Goswami

What a great talk from Swapnil Sayan Saha, on Physics-Aware Tiny Machine Learning! https://lnkd.in/g9g4YWs4 There's exciting times ahead as Physics-aware TinyML is becoming increasingly important in merging the power of machine learning with operating in the physical world. By integrating real-world physics models into ultra-low-power devices, this approach dramatically enhances the accuracy and efficiency of TinyML applications. From improving energy efficiency in smart sensors to enabling more accurate anomaly detection in industrial applications, this approach to machine learning at the edge helps machines better understand their environment - without sacrificing power or performance. As we explore the possibilities, this combination of physics and Edge AI opens up new opportunities in fields like healthcare, environmental monitoring, and autonomous systems. It's a huge leap toward making AI more sustainable and impactful, and allows us to build more useful AI-enabled systems that operate in (and on!) the physical world. #TinyML #EdgeAI #PhysicsAwareAI #AIInnovation #Sustainability #MachineLearning

4

This AI Paper by the National University of Singapore Introduces MambaOut: Streamlining Visual Models for Improved Accuracy

6

Veronica ab hindi mai baat kar sakti hai! 😎 With a neural machine translation model and a text-to-speech model trained on IndicTrans data, Veronica is now multi-lingual! 🗣️ Z by HP Z6G5A running all the multiple models, making Veronica come to life! 🚀 #ai #deeplearning #artificialintelligence #largelanguagemodels #generativeai #genai #coder #coderlife #technology #techdemo

118

13 Comments

Understanding the use of multimodality in RAG based system can help a lot in making business grow and most definitely there could be other uses cases as well. I would like to thank Mr. Andrew Ng for creating this beautiful short course that help understand a lots of things around RAG and its multimodality.

11

1 Comment

We partnered with PyTorch to accelerate its adoption for building AI experiences. Come to #SnowflakeDevDay on June 6 in SF to learn how to accelerate #GenAI performance using native PyTorch

6

I just came across a fascinating paper that helped me crystalize a part of our work at Neon to achieve very high computational efficiencies on our models: switching from NxN to Nx1 convolutions. Drawing on insights from BitNet, which quantized LLMs to binary values, this paper introduces matrix multiplications with matrix-multiplication-free dense layers with ternary weights, offering significant efficiency gains. The performance benefits from pure algorithmic sense are simple to grasp. Matrix multiplication is O(n^3) and, in the best case, O(n^~2.4). Linear operations are O(N), but you will trade that for depth, so the correct value is n_layers * O(N). But that's just the tip of the iceberg. Now let's look at a H100 GPU -- what makes that fast? HBM3 memory and a pile of other tricks that reduce latency to half of an A100. nVidia invested in memory latency and bandwidth because they are a big part of keeping a monster GPU fed. Matrix multiplications cause non-local access and thus add to the complexity needed to overcome those non-local accesses. That is why fast MatMul ops only help at large matrix sizes. Linear operations, on the other hand, are cache-friendly, pre-fetchable, and very friendly for SIMD operations. These hardware advantages allow models to work better not only GPUs but also CPUs and NPUs. These findings are not just theoretical. They have practical implications, demonstrating that a well-designed network can effectively replace the cross-correlative weighing functions provided by matrix multiplication with depth-wise operations. These operations are not only common in tokenization but also find applications in various other areas. This shift towards more architecture and efficiency-aware designs is a promising development, opening up new possibilities for the use of LLMs and generative models on a wider range of devices. For a more in-depth exploration, check out the links: 📜Scalable MatMul-free Language Modeling: https://lnkd.in/gVpMhjdU 📜BitNet: Scaling 1-bit Transformers for Large Language Models: https://lnkd.in/gzFbtduJ 💡Example of MatMul use in CLIP (AI by Hand): https://lnkd.in/gVU6CYhh #ComputationalEfficiency #MachineLearning #EfficiencyAwareDesigns #LLMs #GenerativeModels #ResearchPapers

25

"QUANTUM MACHINE LEARNING" Quantum Machine Learning (QML) combines principles from quantum computing & machine learning. It aims to leverage the power of quantum computing to accelerate and enhance machine learning algorithms. While classical machine learning relies on traditional computers to process large datasets and detect patterns, quantum computing uses quantum bits (qubits) to exploit quantum phenomena such as superposition and entanglement, potentially offering exponential speedups for certain tasks. By harnessing quantum computational resources, QML could revolutionize areas like optimization, data analysis, & artificial intelligence, opening up new possibilities for solving complex problems in science, medicine, finance, & more. Quantum Machine Learning holds significant promise for advancing the field of artificial intelligence by overcoming limitations of classical systems. Even though the technology is still in its early stages, the potential for quantum algorithms to handle large-scale, high-dimensional data more efficiently offers a glimpse into the future of computation. As quantum hardware continues to improve & QML algorithms are further developed, it is likely that this field will play a pivotal role in shaping next-generation AI systems, unlocking capabilities that were previously thought to be unattainable. Check out the conversation with Mr. Gaurang Belekar, Researcher at Trinity College Dublin in Quantum Science & Technology about Deep Learning, Neural Networks, Building ML Algorithms & Deep Learning Frameworks, Quantum Machine Learning, Quantum Hardware, Qubit Technology, & more, on YouTube from the link below: https://lnkd.in/dRY_6XS3

9

2 Comments

Generative AI and Google Gemini Models Excited to share my latest project on Generative AI with Google Gemini Models! I've uploaded the code and examples to my GitHub. You can check it out here: 1. https://lnkd.in/dMkdP32N 2. https://lnkd.in/dXHSxkvA I’m eager to collaborate with like-minded AI enthusiasts, researchers, and professionals who are passionate about pushing the boundaries of generative AI. Let’s work together, share insights, and build something amazing! If you're interested, feel free to connect and contribute. Looking forward to hearing your ideas! Salam Ullah Khan #GenerativeAI #GoogleGemini #AI #GitHub #Collaboration

5

2 Comments

https://lnkd.in/gpPTGKM4 - enjoyed the latest short course from deeplearning.ai and Google. The course is good as a starting point on multi-modal prompts, but expect the model outputs to be a little more professional, like for ex: in the example of 'Interior design' - it would have been great if the model output gave the updated Interior with the new Furniture added, rather than a textual output, saying which Furniture does / does not fit in with the original design. I believe such systems are at use in many of the latest stores like IKEA. But overall a great course.

6

A great post describing our work on Pytorch based AI-trainer fault tolerance using distributed checkpoints. The work is available as open source pytorch module -> Pytorch Distributed Checkpointing. 🎉 🎉 https://lnkd.in/e6dgTuHw

10

Excited to announce that our paper, "Context-Aware SQL Error Correction Using Few-Shot Learning," has been accepted for presentation at the 1st Workshop on GenAI and RAG Systems for Enterprise during the #CIKM2024 Conference! Text-to-SQL is a rapidly growing area. Many companies have large databases but must only rely on data engineers and data scientists to write SQL to extract information for insights. Automated SQL generation bridges the gap between non-technical users and databases, enabling them to retrieve and interact with data without needing deep knowledge of SQL. However, generating accurate SQL is challenging due to the complexity and variability of natural language questions. In our paper, we propose a RAG-based few-shot learning solution in our error correction agent, improving the fix rate from 28.9% to 39.2%. I'd like to thank my co-author Divyansh J. and other members in our text-to-SQL project (William Hicks, Ben Jarmak, Carl Simon Adorf, Jiwei Liu, Ph.D., Charles Blackmon-Luca, Fay Wang, Lucy Zhang, Prakash G., Vibhor Agrawal). You can read the full paper here: https://lnkd.in/gZedqdU2

45

📢Our paper got accepted 😄 CORE-ReID is ranked 1st on "Papers with code" in Unsupervised Domain Adaptation for Re-Identification. Benchmark: https://lnkd.in/gNCmr3P7 Paper: https://lnkd.in/gFtRX7KS

14

𝐅𝐢𝐧𝐞 𝐓𝐮𝐧𝐢𝐧𝐠 𝐔𝐬𝐢𝐧𝐠 𝐅𝐞𝐝𝐞𝐫𝐚𝐭𝐞𝐝 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠: 𝐂𝐨𝐮𝐫𝐬𝐞 𝐓𝐨𝐩𝐢𝐜: Fine-tuning or training a model with private data using federated learning. The result of centralized training of a model on data and federated training a data yields same accuracy but the later preserves privacy. 𝐏𝐫𝐢𝐯𝐚𝐜𝐲 𝐂𝐨𝐧𝐜𝐞𝐫𝐧: Prompts and perplexity analysis can reveal training data. 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Use differential privacy to obscure whether a response comes from actual training data. 𝐓𝐫𝐚𝐝𝐞-𝐨𝐟𝐟: Increased privacy reduces the model's ability to learn detailed patterns, balancing accuracy and privacy. 𝑪𝒐𝒖𝒓𝒔𝒆 𝑳𝒊𝒏𝒌: https://lnkd.in/dGkK8j4Q

13

A very nice (quick 20 minute) introduction to Responsible and Safe AI. Check it out (stream link below). And if interest in knowing more.. join the 4 weekend workshop for tech leaders and architects (Details: https://lnkd.in/gry9a442 )

12

Sharing as this was the first time a publication that I involved attracted 200+ citations. (Not a bad milestone as a non-researcher looking back at my scholar profile in ages 😂 ) Well before the age of Generative-AI, Prof. Shinichi Nakagawa's vision for automation of mapping of research sysnthesis was published as "Research Weaving: Visualizing the Future of Research Synthesis". This includes a vision, discussion and early stage prototyping. Shinichi Nakagawa Neal Haddaway Martin Westgate Rose E. O'Dea Daniel W.A. Noble Malgorzata Lagisz https://lnkd.in/gap96Ysh #ResearchSynthesis #AI #NLP

53

4 Comments