Gautier Berthou

Lately, I've been using Windows and got annoyed by Alt-Tab. So I wrote a tool to let you switch focus with the arrow keys (or the home row). https://lnkd.in/drTX5kUj Also comes with a handful of i3-esque goodies like desktop switching, window swapping, and focus borders. So much nicer than having to constantly reach for the mouse. Would recommend combining with FancyZones (https://lnkd.in/d6hM_fUV) and AltDrag (https://lnkd.in/dyUDGD86) to get a more complete tiling window manager experience.

62

4 kommentarer

🚶‍♀️ Take a quantum walk; they're resistant to noise and oh so peaceful. Today's dispatches from quantum: ⚆ New York University and the Niels Bohr Institute, University of Copenhagen are collaborating to develop superconductor-semiconductor materials for quantum computing ⚆ A study demonstrates the local writing and erasing of light-emitting defects in silicon for integrating quantum emitters in large-scale qubit systems ⚆ Research investigates continuous-time quantum walks for quantum search due to their resilience against noise and dissipation ⚆ A new method for reading quantum states in nitrogen-vacancy centers in diamond reduces experimental time while maintaining high fidelity Plus, the Hyperform consortium, including IDEMIA, is developing quantum-safe cybersecurity solutions, researchers have introduced probabilistic quantum algorithms using the linear combination of unitaries method, the second cohort of startups for the QAI Ventures Accelerator has been announced (congratulations Commutator Studios, MQI - Munich Quantum Instruments, QCentroid, QPerfect, Quantized Technologies Inc., Scenario X, and ZuriQ!), and a new NSW Government report projects a $4.6 billion industry in Australia by 2040, with Sydney as a key hub. We're THE quantum computing newsletter, bringing you daily insights into the latest news & research. With leaders in academia, founders, and subscribers from top companies like Xanadu, SandboxAQ, and Qunnect, you're in good company. Join now at www.thedailyqubit.com and never miss a single qubit. #quantumcomputing #quantumphysics #innovation

10

A new paper is just out from our research group! In collaboration with colleagues at the Institute for Energy Technology and Uppsala University, Øyvind Sommer Klyve investigated when retrofitting of existing wind power plants with PV makes most sense, a highly relevant topic today 😁 In the paper, 128 operational wind power plants in Norway and Sweden from which we had data access were addressed. The effect of retrofitting each of these sites with PV power plants was modelled based on meteorological data with hourly resolution. Cost-effective PV capacities and curtailment levels were determined for each site. Among the key insights from this work are the following: 😁 In 97 of the 128 investigated cases, retrofitting is beneficial based on the assumptions in the paper. These are mostly located in the southern regions of Norway and Sweden. 😁The three most important factors determining this are in order of importance: 1: High mean PV capacity factors (good solar resource) 2: Low mean wind capacity factors (less good wind resource) 3: Strong anti-correlation between the hourly PV and wind power generation The applied framework is an effective screening tool for hybrid power plant development, both for retrofitting and greenfield projects. Our part of the work was performed within the KSP HYDROSUN project funded by The Research Council of Norway. I am extremely grateful for both the funding and the very good collaboration with Uppsala University Øyvind Sommer Klyve - Ville Olkkonen - Magnus Moe Nygård - David Lingfors - Oskar Lindberg

44

6 kommentarer

An innovative methodology of ours (converting NDVI data into vegetation phenology indicators, e.g., the onset of greening (OG) in spring and the onset of dormancy (OD) in the autumn), which is introduced and used in our paper (https://lnkd.in/edsNFcBN), was not highlighted too much as the focus of the paper was not on the methodology. So, I decided to put it in my GitHub repository and to provide instructions and example usage to promote it. Methods introduced in the literature that also use the NDVI data to derive OG and OD assume that bud break for a given pixel occurs at a specific NDVI threshold that is greater than the observed NDVI minimum value in winter and early spring. For example, OG is commonly defined as the day of the year when NDVI reaches the NDVI minimum plus 20% of the NDVI range or the critical NDVI value determined from the long-term average annual curve of NDVI. The same also applies to OD. Here, however, we introduce an innovative method that determines this critical NDVI value by cumulatively plotting NDVI. This allows us to show that the curve consisted of three parts that could be approximated linearly and that correspond to different growing periods. The critical NDVI values were then determined by calculating the intersection of the bisectors of the first two (in the case of OG) or last two (in the case of OD) lines and the cumulative curve rather than an arbitrary criterion of minimum NDVI plus 20% of the NDVI range or long-term average critical NDVI. Any remaining uncertainties in our OD and OG estimates are therefore systematic, which can be important in many research projects where a trend analysis is planned. On the other hand, our introduced method, namely the logistic function derivative of the NDVI (LFD-NDVI) curve method, is an innovative algorithm that could operate independently for each year and pixel. This is important because classical methods, e.g., Piao et al. (2006) https://lnkd.in/e7zygQyV, typically calculate a critical long-term NDVI value for the entire period and then determine the timing for approaching such a critical value in all individual years. The LFD-NDVI method uses the first and second derivatives of the fitted logistic function of the cumulative NDVI curve to determine OG and OD. https://lnkd.in/eeKKemxt

48

1 kommentar

I am excited to share an update on a small project of mine! https://lnkd.in/e28pz7t7 I have been working on creating a flexible Variational AutoEncoder (VAE) that is simple to use and can be easily reconfigured for various use cases. This includes feature reduction, probability capture, or generative models such as diffusion probabilistic models. Additionally, this VAE also offers some quality-of-life features such as non-balanced multi-GPU support and batch estimators. And there's more to come! Stay tuned for some helpful how-to guides and Jupyter notebooks in the pipeline. #AI #DeepLearning #VAE #GenerativeModels #AI #NeuralNetworks

8

2 kommentarer

Arega Getaneh, PhD from DTU Wind and Energy Systems presenting our work “Learning to bid in forward electricity markets using a no-regret algorithm” at #PSCC2024. The article is available at: https://lnkd.in/dGkY9-yD Perhaps, the key takeaway is that, thanks to ex-post aggregate supply-demand curve information being published every day by European NEMOs as a common practice (such as Nord Pool: https://lnkd.in/dN7FMs37), market participants can use already well-established no-regret "full feedback" leaning methods to learn how to strategically bid in electricity markets. By full feedback, it means market participants can calculate ex-post their profit not only for the price bid they placed but also for any other price bids that they could place (but didn't).

143

2 kommentarer

I really don’t understand the fuzz about watermarking output of LLMs. This addresses no problem. One of the major problems is that bad actors can just use any LLM (maybe an own trained one) to generate text for massive automated disinformation campaigns. Instead of relying on good actors doing good, can we think about bad actors? we need to think about the societal shift that might require watermarking for human-made content? And assume everything else is AI ? #trusthworthyAI #AI https://lnkd.in/ettuFfKu

36

2 kommentarer

This week at [ICML] Int'l Conference on Machine Learning catch up with Anders Christensen and Nooshin Mojab at the GRaM Workshop https://lnkd.in/d9b76E-h Anders will be presenting results of our work on “Geometry Fidelity for Spherical Images” https://lnkd.in/dPScMk6y Part of the problem for immersive generative content is that, because of its rare format needs, often we rely on small datasets that hinder training, adding that traditional optimization functions don’t transfer so well to other formats. To achieve better results we must improve the way we evaluate our models. Here we introduce OmniFID to improve evaluation for spherical images. If you can’t make it to ICML, don’t worry we will have a more detailed paper with that and other improvements for spherical images at the main track of the European Conference on Computer Vision. Anders Christensen, Nooshin Mojab, Khushman Patel, Karan Ahuja, Zeynep Akata, Ole Winther, Mar Gonzalez-Franco, Andrea Colaco

22

2 kommentarer

PICUP (Partnership for Integration of Computation into Undergraduate Physics) Fall 2024 Webinar Series is here! Teaching Physics Creatively using Computational Essays Tor Ole B. Odden,  University of Oslo, Norway https://lnkd.in/gfCa2yQU Tuesday December 3 at 2 pm ET, 1 pm CT, 12 pm MT, 11 am PT To get the Zoom link, follow https://lnkd.in/gixydegP Physics is a creative discipline, but the ways we teach physics often don't reflect this creativity. At the Center for Computing in Science Education at the University of Oslo, Norway, we have spent the last 6 years exploring how computation can help students to engage in creative research-like activities early in their physics careers. Using a teaching tool known as a computational essay, we have supported students in conducting and presenting independent computational research projects and studied the effects of these experiences on their views of physics and development of computational literacy. In this talk, I will describe what computational essays are, how we have used them in our teaching, and what we have learned about the importance of creative and agentic work in becoming a physicist. American Association of Physics Teachers - AAPT American Physical Society #ComputationalPhysics #PICUP #ComputationalEssays #UndergradPhysics

8

Our latest publication in MNRAS, led by Louisa Mason, takes technosignature research to new heights, quite literally! Using ALMA data, this study explores the potential for narrowband signal detection at millimeter wavelengths—pushing the boundaries of SETI science into new areas of unknown parameter space. Our SETI team at Manchester continues to grow, driving innovation in the search for extraterrestrial intelligence! See also: https://lnkd.in/e_sQcQyH 🔭 #SETI #Technosignatures #ALMA #Astronomy #Research

136

4 kommentarer

✨ EXCITING NEWS ✨ Our extensive research on the capabilities of BART and MarianMT in the field of error correction has been published as two conference papers at IEEE and a journal paper at JIKM. Our research on these models took place in three stages: 1️⃣ Initial Investigation: We started by investigating the capabilities of BART and MarianMT in error correction by proposing a system that enhances the accuracy of noisy text documents using these deep network language models. The results obtained from the research demonstrated a 35% reduction in WER using the BART language model. 2️⃣ Quantifying Effectiveness: We then quantified the effectiveness of BART and MarianMT in error correction, focusing on spelling, grammar, and mixed categories. While both models reduced erroneous sentences by over 20%, BART proved more effective with spelling errors (24.6%) compared to grammatical errors (8.8%). 3️⃣ Comparative Study Between Error Types: Finally, we examined whether the models' error correction capabilities are affected by the type of error (synthetic vs. natural). BART exhibited consistent performance for both types. We also identified the break-even point for BART, where its error correction capability starts to decline, at approximately 26% synthetic error introduction. I'm incredibly proud of the collaborative effort and the impactful results I achieved with Peeta Basa Pati, S.A. Gandheesh, Gayatri Sanjana Sannala, and Suriya KS. Looking forward to more exciting research in the field! For reference, the mentioned papers are listed below: 1. System for Enhancing Accuracy of Noisy Text using Deep Network Language Models: https://lnkd.in/gUDxdKhb 2. Grammatical versus Spelling Error Correction: An Investigation into the Responsiveness of Transformer-Based Language Models Using BART and MarianMT : https://lnkd.in/gkpYgmVG 3. Comparative Study on Synthetic and Natural Error Analysis with BART & MarianMT: https://lnkd.in/g3ZDCKdU

44

A very interesting аrticle published in Nature about LLMs. https://lnkd.in/dms_2y7e So basically, the more complex the model, the more it pretends to know something rather than actually knowing it. This seems like an interesting amplification of an inherent flaw in LLMs, now scaled up in a way that's worth paying attention to. It will be interesting to see if anyone will manage to mitigate this. #artificial_intelligence #ai #llm #genAI

24

2 kommentarer

Understanding the Temperature parameter in Transformer Models: Balancing Creativity and Coherence. Today I'd like to break down the concept of Temperature in Transformer models. I've noticed that many people think of it as a "magic" setting that adjusts the creativity of generated text. Let's take a closer look and explore how it actually works behind the scenes! Spoiler alert: it's pure math :) When it comes to the output of transformer-based models like GPT, the temperature parameter plays a crucial role in shaping the nature of the generated text. Whether you're aiming for creative, diverse responses or more focused, predictable outcomes, this parameter allows you to fine-tune the behavior of the model. 1. What is the Temperature Parameter? 👉 In simple terms, the temperature parameter is a scaling factor that modifies the model’s raw output scores, known as logits, before they are converted into probabilities through the softmax function. See the attached equation of the Softmax activation function and the temperature parameter (T). By adjusting the temperature, you can control how likely the model is to pick more or less probable tokens when generating text. 2. Impact on Token Generation 👉 Low Temperature (< 1): When the temperature is set below 1, the logits are scaled up, making the model more confident in its token choices. This sharpens the probability distribution, leading to more predictable and deterministic outputs. The model tends to generate common phrases or sequences and might become repetitive or conservative at very low temperatures. 👉 High Temperature (> 1): A temperature greater than 1 flattens the probability distribution, scaling down the logits and introducing more randomness. This encourages the model to explore less probable tokens, resulting in creative, diverse, and sometimes unexpected text. However, the higher the temperature, the more likely it is that the model will produce incoherent or chaotic responses. 👉 Temperature = 1 (Baseline): When the temperature is set to 1, the logits remain unchanged, and the model samples directly from the original probability distribution. This is the default setting, balancing both creativity and coherence in the generated output. I hope this blogpost will help you to better understand this concept! #AI #MachineLearning #GenerativeAI #Transformers #NLP

8

New preprint on the arXiv! https://lnkd.in/dsJyRpU8 We present a novel entangling gate protocol for quantum computing with Rydberg atoms. Using resonant dipole-dipole interactions, our approach boosts gate speed and stability against Rydberg decay errors. This advancement could open new pathways for scalable quantum computing with neutral atoms. Thanks to Stefano Veroni, Giacomo Giudice, Hannes Pichler, and Johannes Zeiher for the nice collaboration between Universität Innsbruck/Institut für Quantenoptik und Quanteninformation (IQOQI) in Innsbruck, Max Planck Institute of Quantum Optics, and planqc.

59

1 kommentar

🚀 Excited to share a series of new papers on passivation methods for nonlinear systems! The compositional properties of passive systems make them ideal for large-scale interconnected systems, such as networked systems and plug-and-play architectures. By enabling systematic passivation of systems with a shortage of passivity, our latest research helps achieve: • Stability guarantees for complex networks. • Tools for integrating heterogeneous components into plug-and-play systems without reanalyzing the entire system. • A foundation for designing robust, scalable control strategies for cyber-physical systems, power networks, and beyond. Our recent results are distributed across two papers - one focused on SISO systems and the second extending to the MIMO case. These works were led by Miel Sharf 📄 1. “A Characterization of All Linear Passivizing Input-Output Transformations of a Passive-Short System: The SISO Case” This work provides a geometric characterization of all linear transformations that passivize SISO systems. It connects passivity theory to the geometry of cones and introduces powerful tools for system design and stability, reminiscent of the Youla parameterization for stabilizing controllers. Published in IEEE Control Systems Letters. We will present this work at #CDC2024 in Milano next week! 🔗 https://lnkd.in/dq8rx9Gk 📄 2. “A Characterization of Passivizing Input-Output Transformations of Nonlinear MIMO Systems” Extending our results from SISO to MIMO systems, this paper presents a complete framework for characterizing passivizing transformations for nonlinear systems with multiple inputs and outputs. It paves the way for solving optimal passivation problems and enhancing applications in networked and cyber-physical systems. Published in IEEE Control Systems Letters (early access). 🔗 https://lnkd.in/d-4__ZCp To learn more about our research, visit the Connect Lab at the Technion: 🔗 https://lnkd.in/dgD3763G Let me know your thoughts, or reach out if you’d like to discuss how these methods can impact your work! #ControlTheory #Passivity #MIMO #SISO #NetworkedSystems #PlugAndPlay #SystemsEngineering #Research

49

1 kommentar

My second second-author-paper https://lnkd.in/d6qP9sVH just got published and has quite some highlights: ⭐ Together with @Tarik Sahin at Bundeswehr University Munich and Alexander Popp we developed and implemented PINNs for Contact Problems. ⭐ The Fischer–Burmeister NCP formulation helps the optimizer with the contact constraints. ⭐ PINNs are explored as fast-to-evaluate surrogate model for varying loads in the Hertzian contact problem. 👍 👎 🤼 And there is also an interesting point for further discussion. In Scientific machine learning papers with PDE applications, one often reads *“L2-Error”*, but in the end this is a l2 vector norm over values at random sampling points, i.e. Monte Carlo integration. Table 9 compares this procedure with proper domain integration, and… the numbers differ! Do you care? Does it matter in your application? Happy to discuss, leave a comment.

37

Checkout our new article on Instance Segmentation of Plant Point Clouds recently published in COMPAG: https://lnkd.in/erDvy9jV In this article we showed the effectiveness of explicitly making the segmentation embedding space more sparse in which different instances are more easily separable.

56

8 kommentarer

Seminar : Ari Paavo SEITSONEN Our next Physics Seminar Series will occur on Tuesday, 7 May, at 3:15 PM at the UM6P campus (Ryad 8, 1st floor). We will welcome Ari Paavo SEITSONEN, École Normale supérieur Title: DFT calculations in interplay with experiments on two dimensional materials and at interfaces: h-BN, graphite. Abstract: The experimental results often require interpretation of the experimental signal that is often indirect. Theoretical modelling is therefore often useful or even crucial in order to disentangle the experimental signal and understand the underlying strucure and physics. In this respect the Density Functional Theory (DFT) calculations have been an invaluable tool, that not only allows the resolution of the underlying geometric structure, but it can used also used to directly model the experimental signal. In this contribution we will highlight some of the recent projects where we have used DFT modelling to help experimentalists in the studies of two dimensional materials supported on metal substrates. In particular we will concentrate on the hexagonal boron nitride, h-BN, a "cousin" of graphene, but with a large band gap ; this enables it being used as an insulating layer between the adsorbates and the underlying substrate, on which the h-BN was grown. Time allowing we will also discuss methods how to perform calculations of different experimental signals, like vibrational Infra-Red (IR) and Raman spectroscopies. Teams Link : https://lnkd.in/e-txQZ9g

20