Daniele Bernardini

New Paper in Awesome Multi-Agent Papers! ⎆ Improving LLM Reasoning with Multi-Agent Tree-of-Thought Validator Agent ⎆ LLM-AGENT-UMF: LLM-BASED AGENT UNIFIED MODELING FRAMEWORK FOR SEAMLESS INTEGRATION OF MULTI ACTIVE/PASSIVE CORE-AGENTS https://buff.ly/3KloCzK

3

Chapters 8 and 9 of "Building Quantum Software" are in MEAP now. Applications of Quantum Fourier Transforms, and Quantum Phase Estimation. All 12 chapters are going into the final review at the end of the week, we are getting closer to the finish. https://mng.bz/x2jY Manning Publications Co. #quantumcomputing #fourier

75

5 Comments

Introducing ft-Q: Improving Vector Compression with Feature-Level Quantization Michelangiolo Mazzeschi #VectorCompression #Quantization #MachineLearning #DeepLearning #FeatureEngineering #AI #DataScience #ModelOptimization #TowardsDataScience #MLTechniques #thepixelmuses https://lnkd.in/et9UrEie

4

The Extreme Theory of Functional Connections (X-TFC) is an innovative approach applying physics-informed neural networks to solve optimal control problems with remarkable precision and efficiency. The paper applies X-TFC to three classic control problems: 1. Feldbaum Problem – optimizing trajectory with feedback control; 2. Minimum Time Orbital Transfer – calculating the path that minimizes travel time between two orbits; 3. Maximum Radius Transfer – finding the trajectory with the largest final radius within a fixed time. How Does X-TFC Work? X-TFC uses a concept of 'constrained expressions' that directly integrates boundary conditions into the calculation, removing the need for additional terms in the loss function. This is achieved through a single-layer neural network trained via the Extreme Learning Machine (ELM) method. ELM leverages random weights and biases, computing only the output weights, which leads to a drastic reduction in training time. To tackle optimal control problems, X-TFC applies Pontryagin’s Maximum Principle, producing a set of Ordinary Differential Equations (ODEs). The solutions to these ODEs represent optimal state and co-state trajectories, calculated through the ELM neural layer within the X-TFC framework. The results are remarkable: Feldbaum Problem: X-TFC achieved a precision with an absolute error of approximately in just 0.006 seconds. Minimum Time Orbital Transfer: Using an indirect approach, X-TFC outperformed methods based on Chebyshev series (as seen in the reference paper), which require around 9 seconds to solve the problem. X-TFC completed the calculation in under 0.03 seconds with superior accuracy, meeting all boundary conditions with near-zero error. Maximum Radius Transfer: This problem requires maximizing the final radius in a fixed heliocentric orbit. X-TFC again achieved high accuracy and maintained exact boundary conditions. In comparative tests, X-TFC outperformed traditional methods and software like GPOPS-II, achieving dramatically shorter calculation times. X-TFC stands out not only for its precision and speed but also for its low sensitivity to nonlinear problems. Unlike deep networks that heavily depend on gradient-based methods, X-TFC uses a single-layer network with fixed weights, making it less prone to initialization errors and optimizing training with ELM. This lightweight, efficient structure not only improves execution time but also reduces computational costs, making it highly promising for real-time control in space trajectories. In the end, X-TFC is a true game-changer for those working at the intersection of AI and applied physics. It seems poised to drive advances in fields like aerospace guidance, fuel optimization, and more, enabling complex, low-latency solutions to become part of our everyday toolkit in advanced engineering applications. #PhysicsInformedAI #OptimalControl #ArtificialIntelligence #XTFCTechnology #PINNs #SpaceTrajectories Link on first comment!

22

2 Comments

0063: Quantum Chemistry Simulation Methods using Quantum Subspace Expansion (QSE) The Quantum Subspace Expansion (QSE) method has garnered attention for its ability to extend the utility of quantum algorithms like the Variational Quantum Eigensolver (VQE). While VQE is effective at finding the ground state energy of a molecular Hamiltonian, it may struggle to capture excited states or provide highly accurate results due to noise and limited quantum resources. This is where QSE proves to be a valuable augmentation. QSE builds on the concept of exploring a subspace of states around the ground state obtained from VQE. By constructing a small subspace spanned by basis vectors—typically generated by applying excitation operators to the VQE solution—QSE allows for more refined energy calculations, including the estimation of excited states. This subspace is diagonalized, resulting in an extended spectrum of eigenvalues that includes both the ground and excited states. The key advantage of QSE is that it leverages the same variational principles as VQE but with a broader subspace, providing a more accurate picture of the electronic structure of molecules. The quantum computer is used to measure additional matrix elements of the Hamiltonian and overlap matrices within the subspace, which are then processed classically to determine eigenvalues corresponding to excited states. QSE offers a more accurate quantum chemistry simulation method, particularly for small and intermediate-sized quantum devices. It mitigates errors by effectively exploring a local region around the VQE solution. As quantum devices scale and noise mitigation techniques improve, QSE is expected to become even more relevant for both ground and excited state calculations. Moreover, QSE has practical applications in areas where understanding excited states is crucial, such as in photochemistry, materials science, and the development of photovoltaic materials. Its ability to expand the utility of early-stage quantum devices makes it a promising candidate in the quest to solve complex molecular problems. By providing access to excited states and improving the precision of ground state calculations, Quantum Subspace Expansion is a significant step forward in the practical application of quantum computing to chemistry. As quantum hardware continues to advance, we can expect QSE to play an increasingly vital role in simulating complex chemical systems with greater accuracy and efficiency. PS: Check the comments for links to the code repository and additional resources. For further insights into Quantum Chemistry, see posts from 0051 on my timeline or visit my GitHub repository (linked in the comments). We previously covered another interesting domain - Quantum Econophysics (Posts 0041 - 0050).

45

6 Comments

🚀 Looking to streamline fine-tuning for models like Llama 3, Mistral, and Mixtral? Our latest blog shows how Anyscale simplifies LLMs fine-tuning with Ray: 🔑 Key takeaways: • Scalable fine-tuning with LLM-Forge • Model Deployment with RayLLM Read more: https://lnkd.in/gQwD4A2U

35

📣 Don't miss Marco Gorelli's talk at #PyConItaly on supporting multiple DataFrame libraries seamlessly.🙌 He'll cover the "just convert to pandas" limitations and how the DataFrame Interchange Protocol and Narwhals offer a better solution for robustness and efficiency. 🚀 https://2024.pycon.it/en #DataFrame #Polars #OpenSource

1

NVIDIA has just released a fine-tuned version of Llama 3.1 70b under the name #Nemotron. "Llama-3.1-Nemotron-70B-Reward is a large language model customized using developed by NVIDIA to predict the quality of LLM generated responses. Specifically, it has trained using a Llama-3.1-70B-Instruct Base on a novel approach combining the strength of Bradley Terry and SteerLM Regression Reward Modelling." It seems it beats GPT-4o and Sonnet 3.5 in the RewardBench benchmark. It also shows excellent results in other areas. It was fine-tuned using RLHF. Is NVIDIA trying to take all the market? More at this link: https://lnkd.in/dWpFJWPB

13

2 Comments

💥💥💥 Alopex: A Computational Framework for Enabling On-Device Function Calls with LLMs Yide Ran, Zhaozhuo Xu, Yuhang Yao, Zijian Hu, Shanshan Han, Han Jin, Alay Dilipbhai Shah, Jipeng Zhang, Dimitris Stripelis, Tong Zhang, Salman Avestimehr, Chaoyang He Abstract The rapid advancement of Large Language Models (LLMs) has led to their increased integration into mobile devices for personalized assistance, which enables LLMs to call external API functions to enhance their performance. However, challenges such as data scarcity, ineffective question formatting, and catastrophic forgetting hinder the development of on-device LLM agents. To tackle these issues, we propose Alopex, a framework that enables precise on-device function calls using the Fox LLM. Alopex introduces a logic-based method for generating high-quality training data and a novel ``description-question-output'' format for fine-tuning, reducing risks of function information leakage. Additionally, a data mixing strategy is used to mitigate catastrophic forgetting, combining function call data with textbook datasets to enhance performance in various tasks. Experimental results show that Alopex improves function call accuracy and significantly reduces catastrophic forgetting, providing a robust solution for integrating function call capabilities into LLMs without manual intervention. 👉 https://lnkd.in/dYJZxSW2 #machinelearning

1

New breakthrough in the area of #quantum #AI, where the company Quantinuum has just released a novel concept for designing a quantum transformer. This approach will embark on a new path forward for the #quantum #computing on natural language processing (#nlp). You could find the paper here: https://lnkd.in/gjvrjq7y

33

🚀 Control Systems in AI 🚀 Control systems often face challenges with high-dimensional, noisy data. The ACDF algorithm, improves the learning of action-dependent latent states, and ensures more accurate representation and control in complex environments. Paper 🔗 - https://lnkd.in/gi82aUqn 💡 Keen on AI breakthroughs? DM us to explore opportunities!

9

💥💥💥 Low-Rank Adaptation for Foundation Models: A Comprehensive Review Menglin Yang, Jialin Chen, Yifei Zhang, Jiahong Liu, Jiasheng Zhang, Qiyao Ma, Harshit Verma, Qianru Zhang, Min Zhou, Irwin King, Rex Ying Abstract The rapid advancement of foundation modelslarge-scale neural networks trained on diverse, extensive datasetshas revolutionized artificial intelligence, enabling unprecedented advancements across domains such as natural language processing, computer vision, and scientific discovery. However, the substantial parameter count of these models, often reaching billions or trillions, poses significant challenges in adapting them to specific downstream tasks. Low-Rank Adaptation (LoRA) has emerged as a highly promising approach for mitigating these challenges, offering a parameter-efficient mechanism to fine-tune foundation models with minimal computational overhead. This survey provides the first comprehensive review of LoRA techniques beyond large Language Models to general foundation models, including recent techniques foundations, emerging frontiers and applications of low-rank adaptation across multiple domains. Finally, this survey discusses key challenges and future research directions in theoretical understanding, scalability, and robustness. This survey serves as a valuable resource for researchers and practitioners working with efficient foundation model adaptation. 👉 https://lnkd.in/dQDkHebU #machinelearning

2

1 Comment

Since token prediction is non-deterministic and history dependent - like many complex systems - targeted sampling of alternative token trajectories is prudent when the quality of the output is judged by the entire predicted sequence and potentially multiple fitness measures. “Rolling” alternate futures off of the same trunk history, acting on the preference analysis of these futures, reinforcing the model based on the outcome, and then repeating the process again faster than real time is the hallmark of dynamic predictive control. How many futures to predict, how many “tokens” into the future, and how to assign the credit in the real time learning process, that’s the art of designing and implementing such systems. Exciting to see the same dynamics playing out in large language models too!

5

https://lnkd.in/eyt9tKfN Interesting variation of non causal approach to teaching structured problems to llms

2

In computational linguistics, researchers are focused on enhancing language models' ability to extract precise information from extensive textual data. Recent studies introduce "retrieval heads," specialized attention mechanisms that selectively focus on critical segments of large texts, improving targeted data retrieval. These heads, integrated into transformer-based models like LLaMA and Yi, significantly enhance accuracy and efficiency, particularly in handling long-context scenarios. The methodology involves thorough experiments across various models, including the Needle-in-a-Haystack test, which embeds specific information to measure precision. Results show models with active retrieval heads outperform those without, with accuracy dropping notably when these heads are masked. The empirical evidence underscores the effectiveness of retrieval heads in improving information retrieval precision and reliability. This research deepens our understanding of attention mechanisms in large-scale text processing, paving the way for more efficient and accurate language models. https://lnkd.in/dNNPvMXw https://lnkd.in/ddbHX6EA Innovation Hacks AI #aiinnovation #generativeai

1

【Script-Based GPU-Ready ELM Development for Continuous Code Integration】 Full article: https://lnkd.in/gFM7gFdQ (Authored by Peter Schwartz, et al., from Oak Ridge National Laboratory, Environmental Sciences Division (USA), etc.) The E3SM Land Model (#ELM) is an integral part of the Energy Exascale Earth System Model (E3SM) framework, simulating the interactions between terrestrial land surfaces and other Earth system components. ELM has been instrumental in understanding hydrologic cycles, biogeophysics, and dynamics of terrestrial ecosystems. To facilitate the automatic porting of individual ELM modules, the researchers have developed a #software tool named SPEL. This work presents a SPEL extension that leverages the version control system (e.g., Git) to autonomous code generation and demonstrate its application to #continuous_code_integration and development of the ELM software system.  

2