Roberto Calandra
Dresden, Sachsen, Deutschland
1505 Follower:innen
500+ Kontakte
Info
My research interest is at the convergence of Machine Learning, Robotics and Touch…
Berufserfahrung
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Ausbildung
-
-
–
Thesis title: An Exploration of Deep Belief Networks toward Adaptive
Learning -
–
Thesis title: Design and Build of a Robotics mobile platform
-
Ehrenamt
Weitere Beiträge entdecken
-
Robert Wille
This week, the #Software for #QuantumComputing community of the German-speaking countries gathered for the Quantum Computing #Workshop of the Gesellschaft für Informatik e.V.. Many thanks to Jörg Lässig for the organization and all speakers/attendees for the great discussions!🙂 Interested in more and/or in joining the (German) #Fachgruppe #Quantencomputing? Read more here: https://fg-qc.gi.de #INFORMATIK24
-
Djordje Zikelic
Looking forward to attending PLDI 2024 in Copenhagen this week, where we will present our work "Equivalence and Similarity Refutation for Probabilistic Programs". Equivalence and similarity are two fundamental relational properties of probabilistic programs that are essential for their correctness both in implementation and in compilation. In this work, we present a new method for static equivalence and similarity refutation which is fully automated and provides formal guarantees on refutation correctness. Joint work with Krishnendu Chatterjee, Ehsan K. and Petr Novotný. paper and code: https://lnkd.in/d-xx-Hm2
-
Johanna Miecznikowski
Grouping the TIGR data for reuse: https://lnkd.in/esDwEx4K. After having found out how the LaRS @ SWISSUbase repository is organized and knowing what kind of data the TIGR corpus of spoken #Italian contains (see what we posted in April: https://lnkd.in/eKUiPN8c), the ShareTIGR team is currently trying to figure out how to group the various available files on the repository. We reached some decisions, which required to anticipate user needs at various levels: Will users wish to download all data or subsets? What will they do with the data and how are they likely to be equipped? This week, we report our reasoning so far as well as some open questions. This blog post concludes our reflection on the overall organization of the language resource we are building. In the next few weeks, we will dive deeper into the question of #transcription and describe in some detail one of the two transcript versions that we intend to provide: transcripts formatted as simple texts optimized for the human eye and for analytical workflows outside an integrated digital multimedia environment. At first sight, producing this kind of traditional transcript might not seem to be a complex endeavor. However, as we will see, when discourse is initially transcribed in a multimedia annotator (in our case: ELAN), extracting a plain text version actually raises quite a few theoretical and technical problems! #FAIRdata #openscience #opendata #openresearchdata #corpuslinguistics #italianoparlato
-
Tilahun M. Getu
The 2024 Nobel Prize in Physics ‘’rewards plagiarism and incorrect attribution in computer science’’ is the most profound allegation by a well-known AI leader, regardless of any rivalry in the AI discipline! I believe this allegation must be independently investigated by fiercely objective seasoned scientists, who are willing to defend the truth at any cost! If true, the Nobel committee must immediately rescind the award, irrespective of its long-followed tradition.
-
Kumo.AI
🚀 Exciting News! 🚀 We are thrilled to announce that Kumo AI's position paper "Relational Deep Learning: Graph Representation Learning on Relational Databases," has been accepted at #ICML 2024! 🎉 Join us in Vienna from July 21-27, where our very own Matthias Fey will be presenting this exciting new research area. 🔗 https://icml.cc/ In collaboration with Stanford University, we introduce Relational Deep Learning (#RDL), a new research area for applying deep learning on relational databases by viewing them as temporal, heterogeneous graphs. This method eliminates the need for manual feature engineering, leading to more accurate models built faster. 🚀 We’ve also developed #RelBench, a benchmark suite to facilitate research in this new area, using diverse datasets from Stack Exchange discussions to Amazon book reviews. 📚 Stay tuned for more updates on our presentation format, and check out our paper here: https://lnkd.in/dk24vbSg #ICML2024 #MachineLearning #AI #GraphNeuralNetworks #RelationalDeepLearning #DataScience #KumoAI #StanfordCollaboration
-
Teodora Popordanoska
I am happy to introduce our paper "Consistent and Asymptotically Unbiased Estimation of Proper Calibration Errors", published at #AISTATS24. ✨ This project is a collaboration with Sebastian Gruber, Aleksei Tiulpin, Florian Büttner and Matthew Blaschko. Every proper score decomposes into two fundamental components - proper calibration error and refinement - utilizing a Bregman divergence. We propose a method that allows consistent, and asymptotically unbiased estimation of all proper calibration errors and refinement terms. In particular, we introduce Kullback-Leibler calibration error, induced by the commonly used cross-entropy loss. As part of our results, we prove a relation between refinement and f-divergences, which implies information monotonicity in neural networks, regardless of which proper scoring rule is optimized. For more details, please check our: ➡ Paper: https://lnkd.in/ereUAvVC ➡ Code: https://lnkd.in/eGgYBVfv We are presenting our poster today from 15:00 to 17:00 @ MR1 #31 😊 #deeplearning #machinelearning #classification #uncertainty #calibration #properscores
-
Dr Bhavin Parekh
Meta's new groundbreaking paper "Better & Faster Large Language Models via Multi-token Prediction" proposes a new approach to training language models that shows significant improvements in performance. By training models to predict multiple future tokens at once, sample efficiency is improved, and inference times are sped up by up to 3x, particularly with larger models and batch sizes. This technique also shows substantial improvement over traditional next-token prediction models on coding tasks and generative benchmarks. As model size increases, the benefits of multi-token prediction become more significant, implying greater improvements for larger models. The approach consists of a common trunk that processes the input sequence to generate a latent representation of the observed context. On top of that, multiple output heads are each responsible for predicting a different future token simultaneously and independently. During training, the model is optimized for predicting each of the future tokens independently, which trains the model to improve its predictions by considering multiple future outcomes at each step. The predictions are generated in parallel across the multiple heads, so this doesn't add any computation overhead. At inference time, the model can use the trained output heads to generate multiple tokens at once, speeding up the process. Using Multi-token prediction maintains performance advantages even when models are trained for multiple epochs, demonstrating robustness and durability of training gains. Read more about this technique in Meta's paper "Better & Faster Large Language Models via Multi-token Prediction" on arxiv.org.
-
Learning Chips Lab
After two years of work from Denis Lebold, Felix Schneider, and Fabian Schlenke under the Lead from Prof. Dr.-Ing. Michael Karagounis and Prof. Dr. Hendrik Wöhrle, we have finally released the OpenEye Hardware Accelerator for DNN Inference. It is based on the Eyeriss V2 Hardware Accelerator from the MIT. Features: * Implemented in pure Verilog * Row stationary dataflow * Computing with compressed weights and input activations * Flexible on-chip network * Scalable and flexible architecture that can be adapted using generics/parameters * Python-based front end for simple simulation https://lnkd.in/e-SBttFW
-
COREnext
📰 COREnext Scientific Publication Alert! 📰 👉 Check out our #research paper titled "#Software-Defined CPU Modes." Authors: Michael Roitzsch, @Till Miemietz, Christian von Elm, Nils Asmussen 💡 This paper delves into the innovative concept of defining CPU modes entirely through software. Our researchers demonstrate how this design could function, highlighting the advantages it offers. By unifying all existing modes under a common design umbrella, this approach enforces a cleaner structure and greater control over exposed functionality. Additionally, the flexibility of #software-defined modes opens up exciting new use cases. 👉 Discover the future of #CPU design and its potential impact on computing! #Research #CPUModes #SoftwareDefined #Innovation #TechResearch #FutureOfComputing 🔗
-
Marcel Walter
🎉 Exciting News! 🚀 We're thrilled to announce the latest #opensource release of #fiction v0.6.0, which is part of the Munich Nanotech Toolkit (#MNT)! 🌟 https://lnkd.in/erDJ3MuE This release is a major milestone in our journey to bring you the most advanced #FCN framework in the world 🌍 Here's a breakdown of what's new: 🐍 Python Bindings: Utilize the flexibility of Python's ecosystem combined with the performance of C++! We now deliver official Python bindings with pre-built wheels up on #PyPI. Get started with a simple pip install mnt.pyfiction 💡 SiDB Simulation: We include the most powerful #SiDB simulation engine to date: #QuickExact. It works hard in the back end to enable temperature-aware and atomic defect-aware simulation, automatic exhaustive gate design, operational domain computation, and more. Plus, all algorithms are now compatible with H-Si(100)-2x1 and H-Si(111)-1x1 lattice orientations. 🛠️ Physical Design: Our goal of bridging the gap between conventional #VLSI and #FCN continues with post-layout optimization and wiring reduction algorithms that significantly reduce required layout area and delay! Additionally, we include SAT-based clock number assignment and atomic defect-aware placement and routing. 🐛 Miscellaneous: We've squashed bugs, improved usability, boosted performance, updated libraries, overhauled documentation, and more! Check out the full release notes for a detailed look at all the changes: https://lnkd.in/eWz2-i_C We're immensely grateful to all contributors who helped make this release possible! 🙌 Especially big shoutouts go to Jan Drewniok, Simon Hofmann, and Samuel S. H. Ng for their hard work and ongoing support! Try out fiction v0.6.0 today and let us know what you think! Your feedback is invaluable as we continue to improve and innovate.
1 Kommentar -
institute for storycraft & information gathering
A pre-print paper posted last month by researchers including Dmitry Kobak, Rita González Márquez, and Jan Lause at the University of Tuebingen and Ágnes Horvát at Northwestern University suggests that the effect of large language models (LLMs) like #ChatGPT on #scientificwriting is “unprecedented in quality and quantity.” And not in a good way. The study analyzed 14.4 million abstracts of English-language biomedical papers in the PubMed database. The researchers found that authors’ use of hundreds of “style” words — notably verbs and adjectives —exceeded the frequency that would be expected if they had not employed LLMs (words such as delves, showcasing, underscores, potential, and crucial). The researchers point to some benefits of #AI-assisted writing: it can save time, help people write more clearly, and aid with translations. But they observe notable problems. For one: LLM-generated texts are “less diverse and novel than human-written text," which “can degrade the quality of scientific writing.” If the introductions to papers on the same topic sound alike and “contain the same set of ideas and references,” the researchers caution, it could become harder for readers to identify innovative work and fake research. And they call for reexamining policies and regulations that govern how AI is used in science. https://lnkd.in/eygU9gPm Bess Winston Victoria White Brad Hamilton Elizabeth Moore Tony Loftis
-
Klaus Schoeffmann
I am very much looking forward to the Keynote by Prof. Mario Döller, who will talk about "Re-Identification in Multi-Object Multi Sensor Tracking Environments" at FAIEMA 2024 (https://www.faiema.org/) in Athens, Greece, in October 2024! Here is the abstract of his talk: The advancement of sensor technologies has significantly enhanced the capability of multi-object tracking systems in various domains, such as surveillance, autonomous driving, and robotics. However, the challenge of re-identifying objects as they move across different sensors and viewpoints remains critical for maintaining accurate and continuous tracking. This talk will delve into the methodologies and algorithms developed for re-identification in multi-object, multi-sensor tracking environments. It will cover the integration of diverse sensor data, including visual, thermal, and radar inputs, to improve object recognition and tracking fidelity. Additionally, the discussion will highlight recent innovations in machine learning and data fusion techniques that address occlusions, appearance changes, and identity switching issues. Practical applications and case studies will be presented to illustrate the effectiveness and potential of these approaches in real-world scenarios. This talk aims to provide insights into the current state of research and development in this field, offering a comprehensive overview for researchers, engineers, and practitioners involved in multi-sensor tracking systems.
1 Kommentar -
Dr. Peter Kahlert
Our Chapter on AI and social sciece methodology is out! https://lnkd.in/drt7DQkr We mapped landscapes of research on, with, and about AI, highlight the associative, inter- and transdisciplinary role of AI, advocating a research that differentiates and is not afraid to get in touch with those technologies. A true team effort of the whole crew! It is part of a German essay collection, but publication of an English, post-ChatGPT updated version is already pending.
-
Vicki L. Hanson, PhD
🚀 Attention all software engineers! ACM has launched the first issue of Proceedings of the ACM on Software Engineering (PACMSE), a Gold Open Access journal dedicated to advancing software engineering research. PACMSE covers a wide range of topics, from design to deployment, aiming to improve software quality. Each issue of the journal comprises a curated selection of papers, which will be announced through publicized calls for papers. For more detail on the submission process, refer to intro in the inaugural issue: https://bit.ly/3YwVBti 📚 #SoftwareEngineering #OpenAccess #ACM
1 Kommentar -
Emiliano Dall'Anese
Preprint alert: we updated our paper on feedback-based safe gradient flow for solving the AC optimal power flow (OPF) problem in distribution systems: https://lnkd.in/guepTTE2. Our method brings together online feedback optimization and safe control, to solve the AC OPF problem in real time with feasibility guarantees.
-
Dražen Brščić
📢 The Full Paper submissions for the 20th Annual ACM/IEEE International Conference on Human-Robot Interaction #HRI2025 (https://lnkd.in/ekyvBitk) are now open! 📢 📅 𝗞𝗲𝘆 𝗗𝗮𝘁𝗲𝘀 - 𝗔𝗯𝘀𝘁𝗿𝗮𝗰𝘁 𝗦𝘂𝗯𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗗𝗲𝗮𝗱𝗹𝗶𝗻𝗲: 𝘚𝘦𝘱𝘵𝘦𝘮𝘣𝘦𝘳 24 (abstract, title, and author details required) - 𝗙𝗶𝗻𝗮𝗹 𝗦𝘂𝗯𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗗𝗲𝗮𝗱𝗹𝗶𝗻𝗲: 𝘖𝘤𝘵𝘰𝘣𝘦𝘳 1 Submitting a 𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦 abstract by September 24 is a must, but you can refine your submission until October 1. 📄 Be sure to carefully read the Full Paper call for papers (https://lnkd.in/e-9S3xZy) for detailed guidelines on preparing your submission. 🔗 The submission site is here: https://lnkd.in/egne5v3d. We’re looking forward to your exciting contributions to #HRI2025! #Robotics #HumanRobotInteraction
1 Kommentar -
Daniel Scherer
Our mission: To enable resource efficient quantum computing! To this end, our research group on quantum compilation at the #FraunhoferIIS has worked hard to provide a new approach to the problem of unitary synthesis: For a given quantum functionality (i.e. a unitary matrix), synthesize a quantum circuit. While standard approaches to synthesis do not scale well with regard to classical computation time, we employed reinforcement learning to automatically discover efficient heuristics for unitary synthesis. Christian Ufrecht gave a talk on our results at the #IEEEQuantumWeek in the session "SYS-COFT: Fault-tolerant Compilation". If you would like to learn more or want to connect, simply give us a ping 😊 Check out the preprint: https://lnkd.in/edK4XVnz This paper introduces a deep reinforcement learning method for synthesizing unitaries into quantum circuits. The goal is to find efficient quantum circuits with minimal gate count or depth, focusing on the challenging Clifford+T gate set. Although the task remains exponentially complex as the number of qubits increases, our approach, using the Gumbel AlphaZero tree-search, efficiently synthesizes circuits for up to five qubits with up to 60 gates. It outperforms existing tools like QuantumCircuitOpt, MIN-T-SYNTH, and Synthetiq, setting a strong benchmark for future unitary synthesis algorithms. Thanks Abhishek Dubey, Sebastian Rietsch and Christian Ufrecht for the hard work! #qce24 #quantum #quantumcomputing #reinforcementlearning #fraunhofer
2 Kommentare
Weitere Mitglieder, die Roberto Calandra heißen
-
Roberto Calandra
Restaurant General Manager
-
Roberto Calandra
--
-
Roberto Calandra
--
-
Roberto Calandra
falegname
Es gibt auf LinkedIn 11 weitere Personen, die Roberto Calandra heißen.
Weitere Mitglieder anzeigen, die Roberto Calandra heißen