Tero Keski-Valkama

Hey guys, if you want a really good random number generator for your usecases, do check this open source Library out! Do tag us if you're using and let us know your usecases!

4

Vanio Markov and Vladimir R. will present at the Quantum Matter conference in Workshop 02 this week. Talk title: "Quantum Time Series Similarity Measures and Quantum Temporal Kernels". "In all structural and predictive task scenarios, the quantum kernels exhibited superior performance compared to their classical counterparts. " The team released a paper lat month on the subject: https://lnkd.in/gwdiWHvu. I didn't realize there is a workshop dedicated to Quantum Simulation Technologies too. It seems to be a really good conference. #QuantumComputing #quantumconf

29

5 comentarios

0056: Electronic Structure Calculations using QAOA Quantum Approximate Optimization Algorithm (QAOA) has gained significant attention due to its versatility in solving combinatorial optimization problems. While originally proposed for classical optimization, QAOA is now being explored for applications in electronic structure calculations, especially in scenarios where classical methods struggle with scalability and efficiency. The use of QAOA in quantum chemistry offers a unique approach to finding ground-state energies of molecular systems, bridging the gap between optimization and quantum simulation. The primary objective is to solve the Schrödinger equation for a molecular system. Traditionally, methods like Hartree-Fock or Post-Hartree-Fock (MP2, CCSD) approximations have been employed to approximate wavefunctions and determine the energy states of electrons. QAOA, however, approaches the problem differently by framing it as an optimization task where the goal is to minimize the energy of the system. This is achieved through the construction of a cost Hamiltonian representing the energy of the system and a mixing Hamiltonian to explore different states. By optimizing the parameters in these Hamiltonians, QAOA aims to minimize the cost function, corresponding to the system’s ground-state energy. The process begins with encoding the problem Hamiltonian, which represents the electronic interactions in the molecule. This Hamiltonian is mapped to qubits using methods like Jordan-Wigner or Bravyi-Kitaev transformations, translating the electronic problem into a quantum optimization problem. QAOA’s ansatz, composed of alternating layers of unitary operators based on the cost and mixing Hamiltonians, is then applied. By varying the parameters of these unitaries, QAOA gradually converges to a minimum energy solution, akin to how traditional variational algorithms operate. QAOA allows for tunable parameter optimization, which can provide more efficient exploration of the solution space compared to rigid classical methods. Additionally, QAOA can be implemented on near-term quantum devices, making it a practical candidate for quantum chemistry in the era of noisy intermediate-scale quantum (NISQ) computers. Unlike some other quantum algorithms that require fault-tolerant quantum computers, QAOA’s iterative and variational nature makes it resilient to certain noise levels. The use of QAOA in electronic structure calculations is expected to evolve as quantum hardware matures. With better quantum processors and more sophisticated optimization techniques, QAOA could outperform classical methods in terms of speed and accuracy, especially for large and complex molecules. Research is already underway to refine the algorithm’s convergence properties and extend its application to broader classes of quantum chemistry problems, potentially revolutionizing fields like drug discovery, materials science, and catalysis. Link to code and resources in comments.

58

11 comentarios

We have published a survey paper on neuro-symbolic AI (collaboration of Jonathan Feldstein (bennu.ai), Paulius Dilkas (University of Toronto), Vaishak Belle (The University of Edinburgh), and Efthymia Tsamoura (Samsung Research)): https://lnkd.in/eFRDzBeY #NeuroSymbolicAI : AI has two main schools of thought - one logic-based and the other neural-network-based. Logic-based models are particularly powerful if the goal is to have a model that can be interpreted. For example, in medical applications, it is useful not only to know which treatment is the most promising option but also to increase trust in decision-making processes. Neural-network based approaches on the other hand tend to lead to better predictions. However, neural networks require **a lot** of data, which is not available in many applications. Neuro-symbolic models combine the two approaches to gather the strength of both and avoid the weaknesses of either. They are particularly well suited for small and medium sized business that might have a very good understanding of the problem (which they can encode in logical rules) but do not have a lot of data. If this sounds like it is of interest to you, check it out and pass it on to people who might benefit from it. About the survey: We kept the survey as a handbook to allow engineers to easily extend neural networks with logical concepts to improve their models without having to re-engineer the models. The survey also provides an in-depth introduction to formal logic to bring engineers up-to-speed.

13

4 comentarios

Argonne has released news about its upcoming AuroraGPT 2T model https://www.anl.gov/aurora see slide deck from Euro-Par 2024 https://lnkd.in/gRVrVM69 04: Evaluating AI as Research Assistant Critical: p45 • Without relevant/rigorous/precise/comprehensive evaluation, there is no way to measure quantitatively the strengths and weaknesses of foundation models • Evaluation is an important tool for comparison (need common benchmarks) • Common benchmarks lead to competition: current LLMs race • Only evaluation can tell us what skills are strong and which ones are weak and need reinforcement. So it’s not only about model ranking. • Not Trivial: • What and how to evaluate AI research assistant capabilities is not established • Completely empirical and based only on available the interface: Q&As • Risk of contamination (cramming the leaderboard)

4

> Sharing resource < Nice paper from Pasqal and EDF team: "Implementing transferable annealing protocols for combinatorial optimisation on neutral atom quantum processors: a case study on smart-charging of electric vehicles" by Lucas Leclerc, Constantin Dalyac, Pascale Bendotti, rodolphe griset, Joseph Mikael, and Loïc Henriet Abstract: In the quantum optimisation paradigm, variational quantum algorithms face challenges with hardware-specific and instance-dependent parameter tuning, which can lead to computational inefficiencies. However, the promising potential of parameter transferability across problem instances with similar local structures has been demonstrated in the context of the Quantum Approximate Optimisation Algorithm. In this paper, we build on these advancements by extending the concept to annealing-based protocols, employing Bayesian optimisation to design robust quasi-adiabatic schedules. Our study reveals that, for Maximum Independent Set problems on graph families with shared geometries, optimal parameters naturally concentrate, enabling efficient transfer from smaller to larger instances. Experimental results on the Orion Alpha platform validate the effectiveness of our approach, scaling to problems with up to 100 qubits. We apply this method to address a smart-charging optimisation problem on a real dataset. These findings highlight a scalable, resource-efficient path for hybrid optimisation strategies applicable in real-world scenarios. Link: https://lnkd.in/euHH485M #quantummachinelearning #quantumcomputing #research #paper

25

1 comentario

This article by Pere Martra explores a structured pruning technique for state-of-the-art models, that uses a GLU architecture, enabling the creation of smaller and more efficient large language models. #Llama #LLM

5

Very insightful work by Anthropic’s interpretability team. And an amazing paper, with outstanding writing and figures. The idea is very simple: interpret LLMs by leveraging sparse autoencoders as surrogate models of the MLP of transformer blocks, which allow one to disambiguate the superposition of features captured by a single neuron. A simple idea, but a very careful and complex execution, as it is often the case in our line of work. The paper goes into many details and provide a large array of insights, although the gist of the implementation remains obfuscated due to the closed source nature of Claude. Too bad, because this is the kind of work that we need to better understand and eventually trust LLMs. This is demonstrated by the authors in the section ‘Influence on Behavior’, where they show that clamping some features to either high or low value during inference is “remarkably effective at modifying model outputs in specific, interpretable ways”. Hopefully this kind of work is going to be replicated and generalized to open-weights models, such that we have new ways to steer their behavior. https://lnkd.in/eVym7f_f #interpretability #xai #explainableai #steerableai #anthropic #claude #anthropic

3

New breakthrough in quantum world: The two lattice based : Gap Short Vector Problem GapSVP and Short Independent Vector Problem SIVP were awarded by #NIST as resilient post #quantum #cryptography #algorithms (LWE learning with errors ) are now vulnerable 🌡 . Mr Yilei Chen just published in April 18, 2024 “Quantum Algorithms for Lattice Problems” to crack these crypto algorithms in polynomial time. Article:

6

New Paper 🎉🎉🎉 "Localization Using Two Different IMU Sensor-Based Dead Reckoning System" Abstract Dead reckoning estimates the current position, speed, and direction of moving objects using known position information. Localization determines an object's location on the map, categorized into human and vehicle localization. Autonomous vehicles rely on accurate vehicle localization for effective task execution. While Global Navigation Satellite System (GNSS) is a popular method, weak or absent signals can pose challenges. This study utilizes Inertial Measurement Unit (IMU) sensors for localization, integrating a second IMU to enhance accuracy. Fusing data from two IMU sensors, a dead reckoning system achieves 1.02 degrees and 1.41 meters errors in rotation and translation with a single IMU, and 1.01 degrees and 1.04 meters with two IMUs, respectively. MPG Veysel ALVER Konya Technical University KTUN Robotics Automation Control Research and Development Laboratory (RACLAB) IEEE Xplore #raclab #mpg #ktunraclab #konyatekniküniversitesi #ktunedu #localization #sensorfusion #robotics #robots #development #computerscience #autonomousvehicles #collisionavoidance #targetdetection #UGV #gnss #LIDAR #camera #navigation Reference: İbrahim TOY, Akif DURDU, and Abdullah Yusefi, “Localization Using Two Different IMU Sensor-Based Dead Reckoning System”, 2024 23rd International Symposium INFOTEH-JAHORINA (INFOTEH), East Sarajevo, Bosnia and Herzegovina, 2024, pp. 1-5, doi: 10.1109/INFOTEH60418.2024.10495999. Link:  https://lnkd.in/djhTtKGK

38

We are happy to announce that our paper "SAUC: Sparsity-Aware Uncertainty Calibration for Spatiotemporal Prediction with Graph Neural Networks" has been accepted by ACM SIGSPATIAL 2024! 📜 Paper link: https://lnkd.in/efgH22K5 📑ArXiv: https://lnkd.in/evvjYd2A This paper addresses the challenge of uncertainty quantification in spatiotemporal deep learning, which is often overlooked in favor of deterministic predictions. Sparse, highly-granular spatiotemporal datasets pose additional difficulties. To tackle this, the authors propose the Sparsity-aware Uncertainty Calibration (SAUC) framework, which adjusts uncertainty for both zero and non-zero values. They first transform deterministic spatiotemporal Graph Neural Networks (ST-GNNs) into probabilistic models and then calibrate them using quantile methods. Experiments show that SAUC improves variance fitting and reduces calibration errors by 20% in sparse datasets like traffic accident and crime predictions, demonstrating its theoretical and practical value. We extend our heartfelt thanks to our esteemed collaborators: Dingyi Zhuang, Bu Yuheng, Guang Wang, Jinhua Zhao! #UncertaintyQuantification #SpatiotemporalData #GraphNeuralNetworks #DeepLearning #ACMSIGSPATIAL #AIResearch #DataScience #SparseData

14

The library that integrates the #data processing library #polars with the database system #duckdb is here: https://lnkd.in/d6ZMab-H Now that #polars and #duckdb have comparable performances, the integrated #python library is a good news. The #data orchestrator platform that integrated the two is Dagster: ✅ https://dagster.io/ Comparison of #polars with other tools: https://lnkd.in/dZncufK5

3

1 comentario

Open Access: In this conceptual study, Franz Riederer & Christoph Schankin postulate the peak occurrence of #clusterheadache attacks at spring & autumn equinoxes is due to an interference with synchronization of pacemaker systems. https://lnkd.in/gdShMP7e

9

1 comentario

🖼️🖼️ MCAFNet: A Multiscale Channel Attention #Fusion #Network for Semantic #Segmentation of Remote Sensing #Images ✍️ Min Yuan et al. 🔗 https://brnw.ch/21wP6Mv

124

🔊 This paper explores #deeplearning techniques for continuous human activity recognition (#HAR) using #mm-wave Doppler #radar, targeting applications in #elderlycare. Two strategies for detecting daily activities were developed: one handling un-equalized activity sequences and another using a gradient-based equalization approach. For classification, dynamic time warping (DTW) and long short-term memory (LSTM) models were employed. The study evaluated feature extraction methods, including pixel-level data, unsupervised encoded features (UnSup-EnLevel), supervised encoded features (Sup-EnLevel), convolutional neural networks (CNN), and principal component analysis (PCA). A novel supervised feature extraction pipeline (Sup-EnLevel-DTW and Sup-EnLevel-LSTM) outperformed state-of-the-art unsupervised methods. Surprisingly, the UnSup-PLevel approach achieved strong results without requiring annotations or frame equalization. This work highlights the potential of mm-wave radar and deep learning for non-invasive, continuous HAR. Title "Deep Learning Techniques for Radar-Based Continuous Human Activity Recognition" By Ruchita Mehta, Sara Sharifzadeh, Vasile Palade, Bo Tan, Alireza Daneshkhah and Yordanka Karayaneva 🔎 https://lnkd.in/gC3sHgFJ

5

1 comentario

Gopal Padinjaruveetil - check this bottom part “theoretically solve NP problems in polynomial time” Quantum AI Foundation supports the "2WQC" project, which seeks practical algorithms, applications, and physical realizations for two-way quantum computers (2WQC) that could theoretically solve NP problems in polynomial time. Kevin Kirby (thought you would find this interesting as well) #quantumai #quantum #2wqc https://www.qaif.org/2wqc

6

6 comentarios