About
Passionate about using statistics & machine learning for solving business critical…
Activity
-
💸 76% of Google Ads budgets are wasted! Surprising, right? An audit of 2,000 accounts found that 61% of ad spend goes to keywords that never…
💸 76% of Google Ads budgets are wasted! Surprising, right? An audit of 2,000 accounts found that 61% of ad spend goes to keywords that never…
Liked by Sakshi Jain
-
If you are even a teeny bit inspired by SpaceX’s achievement, let this be your call to go join a company that’s building more than just software…
If you are even a teeny bit inspired by SpaceX’s achievement, let this be your call to go join a company that’s building more than just software…
Liked by Sakshi Jain
-
This MADE my day! Big shoutout to Josh Penny, Ebele Onyema, and the amazing Hinge team for backing us from day one.🔥 💬 DM me if you or your…
This MADE my day! Big shoutout to Josh Penny, Ebele Onyema, and the amazing Hinge team for backing us from day one.🔥 💬 DM me if you or your…
Liked by Sakshi Jain
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
-
-
Quantum Computing and Information Theory, High Energy Physics
Volunteer Experience
-
Mentor
Spark Program
- 5 months
Education
Mentored a 7th grader who was excited to learn more about computers. She had no experience in coding to begin with and was fascinated by making her own piece of application.
More activity by Sakshi
-
🔥 Build IRL Newsletter #22 🔥 🤔 Make joining harder 💭 our personal “why” 👖 Denim soirée & more! 👉 Read it here: https://lnkd.in/gFrkYDua By…
🔥 Build IRL Newsletter #22 🔥 🤔 Make joining harder 💭 our personal “why” 👖 Denim soirée & more! 👉 Read it here: https://lnkd.in/gFrkYDua By…
Liked by Sakshi Jain
-
This year's Nobel Prizes in Physics and Chemistry speak deeply to my personal journey from PhD in chemical physics to AI professional…
This year's Nobel Prizes in Physics and Chemistry speak deeply to my personal journey from PhD in chemical physics to AI professional…
Liked by Sakshi Jain
-
From the 2021 physics prize awarded in nonlinear dynamics and complex systems, to the 2024 physics prize given to Hopfield network and deep learning,…
From the 2021 physics prize awarded in nonlinear dynamics and complex systems, to the 2024 physics prize given to Hopfield network and deep learning,…
Liked by Sakshi Jain
-
🔄 Pivoting is hard, and we’ve learned that the hard way. For 6 months, we kept pushing in the wrong direction, thinking that if we just worked…
🔄 Pivoting is hard, and we’ve learned that the hard way. For 6 months, we kept pushing in the wrong direction, thinking that if we just worked…
Liked by Sakshi Jain
-
Beyond excited to announce my first-ever collab! I designed the bag I always wish I had for travel and work with Adrian Solgaard and Solgaard Design…
Beyond excited to announce my first-ever collab! I designed the bag I always wish I had for travel and work with Adrian Solgaard and Solgaard Design…
Liked by Sakshi Jain
-
🚨Job alert - come work with me and Winnie Street on ML and cognition. DM if interested. Happy to chat.
🚨Job alert - come work with me and Winnie Street on ML and cognition. DM if interested. Happy to chat.
Liked by Sakshi Jain
-
Over the last 1.5 years of community building in AI it's clear to me that there is much work to do to strengthen and grow a broader civil society…
Over the last 1.5 years of community building in AI it's clear to me that there is much work to do to strengthen and grow a broader civil society…
Liked by Sakshi Jain
Other similar profiles
Explore more posts
-
Dr. Aditya Raj
A recent breakthrough titled "Matrix Multiplication-Free LLMs" demonstrates a huge advancement in the area of Large Language Models (LLMs) by reducing computational costs. The authors have eliminated MatMul operations from LLMs, claiming to 10 times reduction in memory usage and a 25.6% increase in training speed, all while maintaining strong performance at billion-parameter scales. Paper link: https://lnkd.in/ggph8qXc #AI #machinelearning #deeplearning #LLMs
-
Smarajit (Sam) Das
The focus alternatively can be to concentrate on more niche, yet fundamental, challenges in AI, honing LLMs to perform specific tasks, tackling the data challenges inherent in LLMs, and improving transfer learning from English to other languages. There is also critical need for efficiency and the integration of external systems and application through agentic frameworks .. Read more at: https://lnkd.in/gzy7naZz
-
HEMANTH LINGAMGUNTA
HEMANTH LINGAMGUNTA The exponential function ex, nature's master of self-replication, mirrors the explosive growth of AI models, where LLMs, VLMs, and APIs evolve at breakneck speeds, their knowledge compounding like compound interest in the bank of artificial intelligence. The Exponential Function and AI: A Natural Connection The exponential function ex is extraordinary in mathematics - it's the only function that equals its own derivative. This property makes it ideal for modeling growth and change in natural systems. Interestingly, this concept has profound implications in the world of artificial intelligence, particularly in training Large Language Models (LLMs), Vision Language Models (VLMs), and APIs: 1. Growth Modeling: The exponential function's ability to model rapid growth mirrors the exponential increase in model performance as we scale up training data and model size. 2. Learning Rates: In optimization algorithms like gradient descent, learning rates often follow exponential decay schedules, inspired by the exponential function's properties. 3. Activation Functions: The popular softmax function, crucial in neural networks, is derived from the exponential function. 4. Natural Language Processing: The exponential distribution is used in various NLP tasks, including modeling word frequencies and topic distributions. 5. Computer Vision: Exponential functions play a role in image processing techniques and feature extraction algorithms used in VLMs. Just as ex elegantly describes natural phenomena, it also underpins many of the mathematical foundations that make modern AI possible. As we continue to push the boundaries of AI, the elegant simplicity of the exponential function remains at the heart of our most advanced technologies. #ArtificialIntelligence #MachineLearning #Mathematics #DataScience Citations: [1] Exponential function - Wikipedia https://lnkd.in/ghczUuhm [2] Exploring the derivative of the exponential function - Math Insight https://lnkd.in/gJgH-k8e [3] Basics of Reinforcement Learning for LLMs https://lnkd.in/ggTK5x3h [4] The exponential function - Math Insight https://lnkd.in/gHmWtp6R [5] Training Large Language Models (LLMs): Techniques and Best ... https://lnkd.in/gh6sC24Q [6] Training and fine-tuning large language models - Borealis AI https://lnkd.in/g8T-vvrj [7] 6.8: Exponential Growth and Decay - Mathematics LibreTexts https://lnkd.in/gUr4Yig8 [8] Exponential Growth and Decay | College Algebra https://lnkd.in/g9f5HpNK
-
Pavlo Molchanov
Apple released details on new LLMs that will run on IPhone (https://lnkd.in/g64AxJhU). My take-aways below: - 3B model parameters. In-line with recent trends on small language models. For example Phi-3 (3.8B) achieves amazing benchmarks (MMLU of 68.8) - Grouped Query Attention (https://lnkd.in/gsR7MmCa) to reduce the KV cache costs by 8x-32x. Google's Gemma2B (https://lnkd.in/gzaGTZWi) uses a single KV cache head. - 49K vocab size - Shares language modeling head and the embedding matrices. Welcome back GPT1 (2018)! These are the largest matrices, sharing will save around 200M parameters if we assume 4096 hidden dim - Low bit quantization, activation quantization and embedding quantization - Time-to-first-token latency is 0.6 ms per prompt token, and a generation at 30tps - For personalization it uses LoRA and 3.5bit compression. Applied to all transformer block layers (qkv, projection, MLP). DoRA (https://lnkd.in/gM9MPbYs) would work better! - Training in https://lnkd.in/gwx8ZaZQ. - Uses data parallelism, tensor parallelism (TP), sequence parallelism (SP), and Fully Sharded Data Parallel (FSDP). TP and SP probably are not used for the small 3B model - Uses in-house data and public training data - Large LLM as a judge for sample rejection and RLHF for post-training alignment Great report, many details, Apple becomes more open. This makes sense as users will be concerned on their data going to the server or OpenAI. Reminds Tesla Autopilot report (https://lnkd.in/gMpHnpS3) from the last year that is must see.
2 Comments
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Sakshi Jain in United States
-
Sakshi Jain
Finance & Strategy | 2025 MBA Candidate at Foster School of Business & Forté Fellow | Ex-EY, Credit Suisse, Deloitte | CPA
-
Sakshi Jain
-
Sakshi Jain
Financial Analyst @S&P Global | CPA Aspirant | Investment Banking | Power BI Data Visualization | Tableau expert |
-
Sakshi Jain
-
Sakshi Jain
66 others named Sakshi Jain in United States are on LinkedIn
See others named Sakshi Jain