Databiomics

AI's Energy Demand Today, Tomorrow, and in the Future. https://lnkd.in/dTGGeeAP

Dear Network of Friends and Professionals, A new article in the newsletter "Mathematical Functions Describe the Biological World" is now available in hashtag #Newsletter-Issue 003. Title: "Generative AI in Biology: Exploring the Mathematical Functions Behind the Methods" by Emmimal Alexander. Please, if you liked, follow and share. Best regards, Leandro. https://lnkd.in/d8-TPQWZ

🌟 Good and Bad News: AI's EnergyDemand: Today, Tomorrow, and in the Future A major problem with AI technology is the high energy required ⚡ to train the algorithms. This energy requirement is expected to increase significantly, as the energy demand of AI is growing between 26% and 36% annually. By 2028, AI could consume more energy than Iceland in 2021, and this demand is doubling every 100 days, leading to an increase in U.S. data center power consumption that could triple to 390 terawatt-hours by 2030, or about 7.5% of the country's projected energy demand. Scientists are researching solutions 🔍💡to increase the efficiency of algorithms to achieve faster results with less energy. Techniques such as scheduling AI workloads during periods of low demand can reduce energy consumption by 12% to 15%. Centralizing computing tasks in shared infrastructures such as data centers and cloud computing also contributes to significant energy savings. The AI industry is known for its high energy consumption. ⚠️ Large-scale AI systems such as ChatGPT require a lot of computing power and cooling. Microsoft's water consumption increased by 34% from 2021 to 2022 due to higher AI computing requirements. College of California researcher Shaolei Ren reports that ChatGPT and other large-scale language models (LLMs) can consume up to 500 milliliters of water per 20 to 50 user prompts or questions. This result was detailed in a paper published on Arxiv earlier this year 📄 (please see References, Li et al., 2023, October). The International Energy Agency (IEA) predicts that the power consumption of data centers will double from 2022 to 2026, similar to the energy profile of cryptocurrency mining. Google DeepMind's new method, "Joint Example Selection for Training" (JEST), significantly reduces the computing resources and time required for AI training. This approach outperforms current models with up to 13 times fewer iterations and 10 times less computational effort and has the potential to make AI development faster, cheaper and more environmentally friendly. While the energy requirements of AI are rapidly increasing, innovative methods such as Google's JEST could make AI training more efficient and sustainable. #AIandQuantumComputing Integrating AI with quantum computing is crucial for sustainable development. Quantum computing's linear relationship between power and energy use makes AI more efficient with a smaller energy footprint. Achieving this requires government support, industry investment, academic research, and public engagement, ensuring sustainable AI advancements and protecting the planet's health. Source primary: https://lnkd.in/dZFQsNXS References: https://lnkd.in/dZNyfNMb https://lnkd.in/d-8uE9tM https://lnkd.in/df3_W4nE https://lnkd.in/dFAkqKDe https://lnkd.in/dDxetipv JEST metodo: https://lnkd.in/d7D9BS8z https://lnkd.in/dSDs9-JP https://lnkd.in/dZNyfNMb CLICK HERE FOR SIGN UP for Synthesia

Newsletter: #MathematicalFunctions #DescribetheBiologicalWorld https://lnkd.in/dCgXG-GV Video about the Newsletter: https://lnkd.in/dMcDD59p

Mathematic Functions Describe the #BiologicalWorld: Mathematical functions play a central role in the modeling biological processes and computational biology. 🔬🧠📊 #MarkovModels (MM) in Sequence Analysis 🧬 #MM are crucial for sequence analysis in DNA, where the transition probability matrix P determines the likelihood of transition from one nucleotide state to another. The transition probability matrix P can describe a first-order MM: P_ij = P(X_(n+1) = j | X_n = i) where P_ij is the probability of transitioning from state i to state j. For DNA sequences with nucleotides A, C, G, T, the transition matrix generated by alignment might look like: P = [ P_AA | P_AC | P_AG | P_AT |    P_CA | P_CC | P_CG | P_CT |    P_GA | P_GC | P_GG | P_GT |    P_TA | P_TC | P_TG | P_TT ] #Geneexpression In gene expression analysis, the negative binomial distribution (NB) can estimate the occurrence of certain expression patterns, described by the formula: P(X = x) = (n + x - 1 choose x) * p^n * (1 - p)^x The NB distribution gives the probability of x failures prior to obtaining s successes in a sequence of Bernoulli trials, p is the probability of success, the probability of obtaining x failures before achieving the s-th success on trial x+s. # Given values: n = 10 # Number of different tissues k = 7 # Number of successes (expressed genes) p = 0.6 # Probability of success q = 0.4 # Probability of failure (1 - p) P(X = 7) = (10 choose 7) * (0.6)^7 * (0.4)^3 Simplification how the calculation is done. https://lnkd.in/dpfEdQ-r... Exponential functions model cell division where growth rates are constant. The function N(t)=No e^rt . It projects the number of cells over time at a constant division rate. #Embedding Embedding methods such as one-hot encoding or word2vec (created by google team) convert categorical biological sequences into numerical vectors, allowing the use of deep learning (by transformer: Softmax, Sigmoid, ReLU, Tahn, Adam, etc.,) and machine learning functions (Binary Cross-Entropy Loss, Categorical Cross-Entropy Loss, Gradient Descent, Linear Kernel and others). Overall, these mathematical frameworks contribute significantly to improving our understanding and capabilities in biological sciences. They enable accurate predictions and analyses that improve both theoretical and practical approaches. There are many others powerful functions applied in Biology. The integration of advanced AI (deep learning and machine learning) opens new frontiers and applications. We often only learn to appreciate and love mathematics when we reach more advanced stages of our academic journey. The marvel of mathematical function applications then becomes clear. Reference: https://lnkd.in/d7H7Ae3y https://lnkd.in/diCY4YiJ https://lnkd.in/dau4q3Wa https://lnkd.in/dxz3sEkk https://lnkd.in/d9zQi8H7 Video: Prof. Dr. Thomas Garrity - Author of the book: Garrity, Thomas A. (2004). All the Mathematics You Missed. Cambridge University Press.

The sentence"I know a guy who knows another guy" can be compared to the way neural networks work, in particular the way information is processed across different layers. This analogy can also be applied to teamwork in companies, where individual contributions lead to a common goal, similar to the way neural networks work in deep learning frameworks. So, how exactly do neural networks function? 🤔 Brief explanation: Neural networks consist of layers of neurons, where the output of each neuron is based on the inputs of the previous layers, which are processed through weights and activation functions🧠. This is similar to a person using information from a connection (e.g. knowing a man who knows another man) to accomplish something more meaningful. The connections in neural networks enable complex pattern recognition and decision making, similar to how different team skills and personalities can work together to solve problems. 🌟 https://lnkd.in/dP-Sue9f Source of image: https://lnkd.in/dEkwu5jj... For a more scientific look : https://lnkd.in/dAvyj-aJ

How do we define dimensions? How do our brains comprehend the physical world and reality? Our minds only scratch the surface of knowledge, and we have difficulty understanding the world beyond three dimensions, which limits our understanding and models of how the physical world operates. For this reason, data science, AI, and generative AI are crucial for exploring fundamental questions in the nature, crossing new frontiers of knowledge and applications in various fields such as physics, biology, chemistry, and other sciences. Reference about dimensin here: https://lnkd.in/gce7NHFv

AI agents - They will bridge the gap between the virtual and reality. https://lnkd.in/eKSZqM29

The next Frontiers of AI systems (cognitive AI neural networks: "digital brains") and agents ("digital mini brains"), article interesting, published in the Frontiers by Wang et al., 2023: https://lnkd.in/etATr2Zi

The image you see below is a symbolic artistic representation of the collaboration between artificial intelligence (AI) 🤖 and scientific progress 🧬🔬. It depicts a humanoid robot hand 🦾 passing a luminous sphere, symbolizing knowledge, to a human hand 👋, surrounded by scientific symbols, DNA strands, mathematical equations, and elements of futuristic technology. This composition is a powerful metaphor for the growing intersection between AI and various scientific disciplines. AI has played a crucial role in advancing human knowledge 🧠, especially in recent decades. As Russell and Norvig (2016) point out in "Artificial Intelligence: A Modern Approach", AI is more than a tool for automating tasks (voice recognition systems, inventory control, surveillance systems, language translation, robots and search systems); it's a means to expand our understanding of complexities in various fields, from molecular biology to astrophysics 🌌. A specific example is the application of AI in genomics, highlighted by Lecun, Bengio, and Hinton (2015) in "Deep Learning". Through deep learning, AI has facilitated faster and more accurate analyses of large genetic data sets, driving discoveries in areas like personalized medicine, drug discovery and molecular biology 🧪. In "Life 3.0: Being Human in the Age of Artificial Intelligence" (2017), Tegmark discusses how AI offers new perspectives and solutions to age-old scientific challenges. He argues that AI not only solves existing problems but also paves the way for new questions and discoveries, reshaping the future of scientific research 🔍. Thus, the image captures the essence of the fusion between AI technology and the human quest for knowledge. AI, as a powerful tool, is redefining the boundaries of what's possible in science. References: Russell, S. J., & Norvig, P. (2016). Artificial Intelligence: A Modern Approach. Prentice Hall. Lecun, Y., Bengio, Y., & Hinton, G. (2015). Deep Learning. Nature. Tegmark, M. (2017). Life 3.0: Being Human in the Age of Artificial Intelligence. Knopf.ces: Russell, S. J., & Norvig, P. (2016). Artificial Intelligence: A Modern Approach. Prentice Hall. Lecun, Y., Bengio, Y., & Hinton, G. (2015). Deep Learning. Nature. Tegmark, M. (2017). Life 3.0: Being Human in the Age of Artificial Intelligence. Knopf.