Search

Reinforcement learning (RL) is a promising approach for aligning large language models (LLMs) knowledge with sequential decision-making tasks. However, few studies have thoroughly investigated the impact on LLM agents capabilities of fine-tuning them with RL in a specific environment. In this paper, we propose a novel framework to analyze the sensitivity of LLMs to prompt formulations following RL training in a textual environment. Our findings reveal that the performance of LLMs degrades when faced with prompt formulations different from those used during the RL training phase. Besides, we analyze the source of this sensitivity by examining the model's internal representations and salient tokens. Finally, we propose to use a contrastive loss to mitigate this sensitivity and improve the robustness and generalization capabilities of LLMs.

Fully connected Graph Transformers (GT) have rapidly become prominent in the static graph community as an alternative to Message-Passing models, which suffer from a lack of expressivity, oversquashing, and under-reaching. However, in a dynamic context, by interconnecting all nodes at multiple snapshots with self-attention, GT loose both structural and temporal information. In this work, we introduce Supra-LAplacian encoding for spatio-temporal TransformErs (SLATE), a new spatio-temporal encoding to leverage the GT architecture while keeping spatio-temporal information. Specifically, we transform Discrete Time Dynamic Graphs into multi-layer graphs and take advantage of the spectral properties of their associated supra-Laplacian matrix. Our second contribution explicitly model nodes' pairwise relationships with a cross-attention mechanism, providing an accurate edge representation for dynamic link prediction. SLATE outperforms numerous state-of-the-art methods based on Message-Passing Graph Neural Networks combined with recurrent models (e.g LSTM), and Dynamic Graph Transformers, on 9 datasets. Code is available at: github.com/ykrmm/SLATE.

The first achievement of highly radiating plasmas in negative triangularity is shown with an operational space featuring high core radiation at high Greenwald fraction obtained with the injection of reactor-relevant seeded gases. These negative triangularity (NT) shape diverted discharges reach high values of normalized plasma pressure (BetaN > 2) at high radiation fraction with no ELMs. We demonstrate that as long as the impurity level in the core is kept low to avoid excessive fuel dilution and impurity accumulation, integration of NT configuration with high radiation fraction not only is achievable but it can lead to confinement improvement with stabilization effects originating from collisionality, ExB shear and profiles changes due to impurity radiation cooling. The underlying physics mechanism is robust and holds for a variety of impurity species. The absence of the requirement to stay in H-mode translates in a higher core radiation fraction potentially allowed in NT shape effectively mitigating the power exhaust issue. The results presented here demonstrate a path to high performance, ELM free and highly radiative regime with rector-relevant seeding gases making this regime a potential new scenario for reactor operation.

The Exame Nacional do Ensino M\'edio (ENEM) is a pivotal test for Brazilian students, required for admission to a significant number of universities in Brazil. The test consists of four objective high-school level tests on Math, Humanities, Natural Sciences and Languages, and one writing essay. Students' answers to the test and to the accompanying socioeconomic status questionnaire are made public every year (albeit anonymized) due to transparency policies from the Brazilian Government. In the context of large language models (LLMs), these data lend themselves nicely to comparing different groups of humans with AI, as we can have access to human and machine answer distributions. We leverage these characteristics of the ENEM dataset and compare GPT-3.5 and 4, and MariTalk, a model trained using Portuguese data, to humans, aiming to ascertain how their answers relate to real societal groups and what that may reveal about the model biases. We divide the human groups by using socioeconomic status (SES), and compare their answer distribution with LLMs for each question and for the essay. We find no significant biases when comparing LLM performance to humans on the multiple-choice Brazilian Portuguese tests, as the distance between model and human answers is mostly determined by the human accuracy. A similar conclusion is found by looking at the generated text as, when analyzing the essays, we observe that human and LLM essays differ in a few key factors, one being the choice of words where model essays were easily separable from human ones. The texts also differ syntactically, with LLM generated essays exhibiting, on average, smaller sentences and less thought units, among other differences. These results suggest that, for Brazilian Portuguese in the ENEM context, LLM outputs represent no group of humans, being significantly different from the answers from Brazilian students across all tests., Comment: Accepted at the Seventh AAAI/ACM Conference on AI, Ethics and Society (AIES 2024). 14 pages, 4 figures

This paper studies the concept of stable perturbation $B\in\mathbb{C}^{n\times n}$ for the core-EP inverse of a matrix $A\in\mathbb{C}^{n\times n}$ with index $k$. For a given stable perturbation $B$ of $A$, explicit expressions of its core-EP inverse $B^{\scriptsize\textcircled{\tiny $\dagger$}}$ and its projection at zero $B^\pi$ are presented. Then, the perturbation bounds of $\parallel B^{\scriptsize\textcircled{\tiny $\dagger$}}-A^{\scriptsize\textcircled{\tiny $\dagger$}}\parallel/\parallel A^{\scriptsize\textcircled{\tiny $\dagger$}}\parallel$ and $\parallel B^{\pi}-A^{\pi}\parallel$ are given provided that $B$ is a stable perturbation of $A$. In addition, we investigate the concept of acute perturbation of $A$. We give a perturbation analysis with respect to core-EP inverses. We provide a condition under which the acute perturbation coincides with the stable perturbation for core-EP inverses., Comment: NO

THz amplitude modulators and switches are considered to be the main building blocks of future THz communication systems. Despite rapid progress, modulation and switching devices in this electromagnetic spectrum lag far behind other frequency ranges. Currently, THz modu-lators face major challenges in consistently producing high modulations depths over large frequency bands. Moreover, a convenient integration for practical applications requires that the modulation/switching properties can be electrically controlled. Devices fulfilling all these con-ditions remain to be demonstrated. In this work we show that W-doped VO2 films grown by direct-current magnetron sputtering can be efficiently used for the development reliable, large-area, broadband THz waves modulators. We demonstrate that W doping not only permits to tune the insulator to metal transition (IMT) temperature of VO2, but also, most importantly, to control the topology of the electrically activated transition. In situ / operando X-ray diffraction and Raman spectroscopy characterizations of the devices, coupled with standard resistivi-ty measurements and time-domain THz spectroscopy, unambiguously demonstrate that the changes in the spatial distribution of the IMT is due to structural distortions induced by W doping.

Dynamic link prediction is a critical task in the analysis of evolving networks, with applications ranging from recommender systems to economic exchanges. However, the concept of the temporal receptive field, which refers to the temporal context that models use for making predictions, has been largely overlooked and insufficiently analyzed in existing research. In this study, we present a comprehensive analysis of the temporal receptive field in dynamic graph learning. By examining multiple datasets and models, we formalize the role of temporal receptive field and highlight their crucial influence on predictive accuracy. Our results demonstrate that appropriately chosen temporal receptive field can significantly enhance model performance, while for some models, overly large windows may introduce noise and reduce accuracy. We conduct extensive benchmarking to validate our findings, ensuring that all experiments are fully reproducible. Code is available at https://github.com/ykrmm/BenchmarkTW .

This paper provides some new characterizations of the diamond partial order for rectangular matrices by using properties of inner inverses, minus order, and SVD decompositions. In addition, the recently introduced 1MP generalized inverse and its dual are used to characterize the diamond partial order as a ${\cal G}$-based one. Finally, the one-sided star partial order is investigated by using 1MP- and MP1-inverses., Comment: 12 pages

Graph Neural Networks (GNNs), especially message-passing-based models, have become prominent in top-k recommendation tasks, outperforming matrix factorization models due to their ability to efficiently aggregate information from a broader context. Although GNNs are evaluated with ranking-based metrics, e.g NDCG@k and Recall@k, they remain largely trained with proxy losses, e.g the BPR loss. In this work we explore the use of ranking loss functions to directly optimize the evaluation metrics, an area not extensively investigated in the GNN community for collaborative filtering. We take advantage of smooth approximations of the rank to facilitate end-to-end training of GNNs and propose a Personalized PageRank-based negative sampling strategy tailored for ranking loss functions. Moreover, we extend the evaluation of GNN models for top-k recommendation tasks with an inductive user-centric protocol, providing a more accurate reflection of real-world applications. Our proposed method significantly outperforms the standard BPR loss and more advanced losses across four datasets and four recent GNN architectures while also exhibiting faster training. Demonstrating the potential of ranking loss functions in improving GNN training for collaborative filtering tasks.

Applying reinforcement learning (RL) to real-world applications requires addressing a trade-off between asymptotic performance, sample efficiency, and inference time. In this work, we demonstrate how to address this triple challenge by leveraging partial physical knowledge about the system dynamics. Our approach involves learning a physics-informed model to boost sample efficiency and generating imaginary trajectories from this model to learn a model-free policy and Q-function. Furthermore, we propose a hybrid planning strategy, combining the learned policy and Q-function with the learned model to enhance time efficiency in planning. Through practical demonstrations, we illustrate that our method improves the compromise between sample efficiency, time efficiency, and performance over state-of-the-art methods.

In this paper, various properties of core-EP matrices are investigated. We introduce the MPDMP matrix associated with $A$ and by means of it, some properties and equivalent conditions of core-EP matrices can be obtained. Also, properties of MPD, DMP, and CMP inverses are studied and we prove that in the class of core-EP matrices, DMP, MPD, and Drazin inverses are the same. Moreover, DMP and MPD binary relation orders are introduced and the relationship between these orders and other binary relation orders are considered., Comment: 20 pages

Prompt learning has been widely adopted to efficiently adapt vision-language models (VLMs), e.g. CLIP, for few-shot image classification. Despite their success, most prompt learning methods trade-off between classification accuracy and robustness, e.g. in domain generalization or out-of-distribution (OOD) detection. In this work, we introduce Global-Local Prompts (GalLoP), a new prompt learning method that learns multiple diverse prompts leveraging both global and local visual features. The training of the local prompts relies on local features with an enhanced vision-text alignment. To focus only on pertinent features, this local alignment is coupled with a sparsity strategy in the selection of the local features. We enforce diversity on the set of prompts using a new ``prompt dropout'' technique and a multiscale strategy on the local prompts. GalLoP outperforms previous prompt learning methods on accuracy on eleven datasets in different few shots settings and with various backbones. Furthermore, GalLoP shows strong robustness performances in both domain generalization and OOD detection, even outperforming dedicated OOD detection methods. Code and instructions to reproduce our results will be open-sourced., Comment: To be published at ECCV 2024

Foundation models have emerged as powerful tools across various domains including language, vision, and multimodal tasks. While prior works have addressed unsupervised image segmentation, they significantly lag behind supervised models. In this paper, we use a diffusion UNet encoder as a foundation vision encoder and introduce DiffCut, an unsupervised zero-shot segmentation method that solely harnesses the output features from the final self-attention block. Through extensive experimentation, we demonstrate that the utilization of these diffusion features in a graph based segmentation algorithm, significantly outperforms previous state-of-the-art methods on zero-shot segmentation. Specifically, we leverage a recursive Normalized Cut algorithm that softly regulates the granularity of detected objects and produces well-defined segmentation maps that precisely capture intricate image details. Our work highlights the remarkably accurate semantic knowledge embedded within diffusion UNet encoders that could then serve as foundation vision encoders for downstream tasks. Project page at https://diffcut-segmentation.github.io, Comment: NeurIPS 2024. Project page at https://diffcut-segmentation.github.io. Code at https://github.com/PaulCouairon/DiffCut

Tokamak plasmas with strong negative triangularity (NT) shaping typically exhibit fundamentally different edge behavior than conventional L-mode or H-mode plasmas. Over the entire DIII-D database, plasmas with sufficiently negative triangularity are found to be inherently free of edge localized modes (ELMs), even at injected powers well above the predicted L-H power threshold. A critical triangularly ($\delta_\mathrm{crit}\simeq-0.15$), consistent with inherently ELM-free operation is identified, beyond which access to the second stability region for infinite-$n$ ballooning modes closes on DIII-D. It is also possible to close access to this region, and thereby prevent an H-mode transition, at weaker average triangularities ($\delta\lesssim\delta_\mathrm{crit}$) provided that at least one of the two x-points is still sufficiently negative. Enhanced low field side magnetic fluctuations during ELM-free operation are consistent with additional turbulence limiting the NT edge gradient. Despite the reduced upper limit on the pressure gradient imposed by ballooning stability, NT plasmas are able to support small pedestals and are typically characterized by an enhancement of edge pressure gradients beyond those found in traditional L-mode plasmas. Further, the pressure gradient inside of this small pedestal is unusually steep, allowing access to high core performance that is competitive with other ELM-free regimes previously achieved on DIII-D. Since ELM-free operation in NT is linked directly to the magnetic geometry, NT fusion pilot plants are predicted to maintain advantageous edge conditions even in burning plasma regimes, potentially eliminating reactor core-integration issues caused by ELMs.

Exposure to environmental factors is generally expected to cause degradation in perovskite films and solar cells. Herein, we show that films with certain defect profiles can display the opposite effect, healing upon exposure to oxygen under illumination. We tune the iodine content of methylammonium lead triiodide perovskite from understoichiometric to overstoichiometric and expose them to oxygen and light prior to the addition of the top layers of the device, thereby examining the defect dependence of their photooxidative response in the absence of storage-related chemical processes. The contrast between the photovoltaic properties of the cells with different defects is stark. Understoichiometric samples indeed degrade, demonstrating performance at 33% of their untreated counterparts, while stoichiometric samples maintain their performance levels. Surprisingly, overstoichiometric samples, which show low current density and strong reverse hysteresis when untreated, heal to maximum performance levels (the same as untreated, stoichiometric samples) upon the photooxidative treatment. A similar, albeit smaller-scale, effect is observed for triple cation and methylammonium-free compositions, demonstrating the general application of this treatment to state-of-the-art compositions. We examine the reasons behind this response by a suite of characterization techniques, finding that the performance changes coincide with microstructural decay at the crystal surface, reorientation of the bulk crystal structure for the understoichiometric cells, and a decrease in the iodine-to-lead ratio of all films. These results indicate that defect engineering is a powerful tool to manipulate the stability of perovskite solar cells.

The core-EP and BT inverses for rectangular matrices were studied recently in the literature. The main aim of this paper is to unify both concepts by means of a new kind of generalized inverse called $W$-weighted $q$-BT inverse. We analyze its existence and uniqueness by considering an adequate matrix system. Basic properties and some interesting characterizations are proved for this new weighted generalized inverse. Also, we give a canonical form of the $W$-weighted $q$-BT inverse by means of the weighted core-EP decomposition.

Materials characterization using electron backscatter diffraction (EBSD) requires indexing the orientation of the measured region from Kikuchi patterns. The quality of Kikuchi patterns can degrade due to pattern overlaps arising from two or more orientations, in the presence of defects or grain boundaries. In this work we employ constrained non-negative matrix factorization to segment a microstructure with small grain misorientations,< 1 degree, and predict the amount of pattern overlap. First we implement the method on mixed simulated patterns - that replicates a pattern overlap scenario, and demonstrate the resolution limit of pattern mixing or factorization resolution using a weight metric. Subsequently, we segment a single-crystal dendritic microstructure and compare the results with high resolution EBSD. By utilizing weight metrics across a low angle grain boundary we demonstrate how very small misorientations/low-angle grain boundaries can be resolved at a pixel level. Our approach constitutes a versatile and robust tool, complementing other fast indexing methods for microstructure characterization., Comment: 22 pages, 7 figures

In this work, we study the out-of-distribution (OOD) detection problem through the use of the feature space of a pre-trained deep classifier. We show that learning the density of in-distribution (ID) features with an energy-based models (EBM) leads to competitive detection results. However, we found that the non-mixing of MCMC sampling during the EBM's training undermines its detection performance. To overcome this an energy-based correction of a mixture of class-conditional Gaussian distributions. We obtains favorable results when compared to a strong baseline like the KNN detector on the CIFAR-10/CIFAR-100 OOD detection benchmarks., Comment: NeurIPS ML Safety Workshop (2022)

While a randomized control trial is considered the gold standard for estimating causal treatment effects, there are many research settings in which randomization is infeasible or unethical. In such cases, researchers rely on analytical methods for observational data to explore causal relationships. Difference-in-differences (DID) is one such method that, most commonly, estimates a difference in some mean outcome in a group before and after the implementation of an intervention or policy and compares this with a control group followed over the same time (i.e., a group that did not implement the intervention or policy). Although DID modeling approaches have been gaining popularity in public health research, the majority of these approaches and their extensions are developed and shared within the economics literature. While extensions of DID modeling approaches may be straightforward to apply to observational data in any field, the complexities and assumptions involved in newer approaches are often misunderstood. In this paper, we focus on recent extensions of the DID method and their relationships to linear models in the setting of staggered treatment adoption over multiple years. We detail the identification and estimation of the average treatment effect among the treated using potential outcomes notation, highlighting the assumptions necessary to produce valid estimates. These concepts are described within the context of Medicaid expansion and retention in care among people living with HIV (PWH) in the United States. While each DID approach is potentially valid, understanding their different assumptions and choosing an appropriate method can have important implications for policy-makers, funders, and public health as a whole.

This paper introduces new classes of generalized inverses for square matrices named GD1, and the dual, called 1GD inverse. In addition, we discuss a few characterizations and representations of these inverses. The explicit expressions of these inverses have been established via core-nilpotent decomposition. Further, we introduce a binary relation for GD1 inverse and 1GD inverse, along with a few derived properties., Comment: 16 pages, Linear and Multilinear Algebra (2024)

A square complex matrix $A$ is called (skew) $J$-Hamiltonian if $AJ$ is (skew) hermitian where $J$ is a real normal matrix such that $J^2=-I$, where $I$ is the identity matrix. In this paper, we solve the Procrustes problem to find normal (skew) $J$-Hamiltonian solutions for the inverse eigenvalue problem. In addition, a similar problem is investigated for normal $J$-symplectic matrices., Comment: 25 pages

Deep learning enables automatic and robust extraction of cardiac function descriptors from echocardiographic sequences, such as ejection fraction or strain. These descriptors provide fine-grained information that physicians consider, in conjunction with more global variables from the clinical record, to assess patients' condition. Drawing on novel transformer models applied to tabular data, we propose a method that considers all descriptors extracted from medical records and echocardiograms to learn the representation of a cardiovascular pathology with a difficult-to-characterize continuum, namely hypertension. Our method first projects each variable into its own representation space using modality-specific approaches. These standardized representations of multimodal data are then fed to a transformer encoder, which learns to merge them into a comprehensive representation of the patient through the task of predicting a clinical rating. This stratification task is formulated as an ordinal classification to enforce a pathological continuum in the representation space. We observe the major trends along this continuum on a cohort of 239 hypertensive patients, providing unprecedented details in the description of hypertension's impact on various cardiac function descriptors. Our analysis shows that i) the XTab foundation model's architecture allows to reach outstanding performance (98% AUROC) even with limited data (less than 200 training samples), ii) stratification across the population is reproducible between trainings (within 3.6% MAE), and iii) patterns emerge in descriptors, some of which align with established physiological knowledge about hypertension, while others could pave the way for a more comprehensive understanding of this pathology., Comment: 12 pages + 2 pages of supplementary material, submitted to IEEE journal

We explore the use of large language models (LLMs) for music generation using a retrieval system to select relevant examples. We find promising initial results for music generation in a dialogue with the user, especially considering the ease with which such a system can be implemented. The code is available online., Comment: LBD @ ISMIR 2023

Inter-modal image registration (IMIR) and image segmentation with abdominal Ultrasound (US) data has many important clinical applications, including image-guided surgery, automatic organ measurement and robotic navigation. However, research is severely limited by the lack of public datasets. We propose TRUSTED (the Tridimensional Renal Ultra Sound TomodEnsitometrie Dataset), comprising paired transabdominal 3DUS and CT kidney images from 48 human patients (96 kidneys), including segmentation, and anatomical landmark annotations by two experienced radiographers. Inter-rater segmentation agreement was over 94 (Dice score), and gold-standard segmentations were generated using the STAPLE algorithm. Seven anatomical landmarks were annotated, important for IMIR systems development and evaluation. To validate the dataset's utility, 5 competitive Deep Learning models for automatic kidney segmentation were benchmarked, yielding average DICE scores from 83.2% to 89.1% for CT, and 61.9% to 79.4% for US images. Three IMIR methods were benchmarked, and Coherent Point Drift performed best with an average Target Registration Error of 4.53mm. The TRUSTED dataset may be used freely researchers to develop and validate new segmentation and IMIR methods., Comment: Alexandre Hostettler, and Toby Collins share last authorship

In image retrieval, standard evaluation metrics rely on score ranking, \eg average precision (AP), recall at k (R@k), normalized discounted cumulative gain (NDCG). In this work we introduce a general framework for robust and decomposable rank losses optimization. It addresses two major challenges for end-to-end training of deep neural networks with rank losses: non-differentiability and non-decomposability. Firstly we propose a general surrogate for ranking operator, SupRank, that is amenable to stochastic gradient descent. It provides an upperbound for rank losses and ensures robust training. Secondly, we use a simple yet effective loss function to reduce the decomposability gap between the averaged batch approximation of ranking losses and their values on the whole training set. We apply our framework to two standard metrics for image retrieval: AP and R@k. Additionally we apply our framework to hierarchical image retrieval. We introduce an extension of AP, the hierarchical average precision $\mathcal{H}$-AP, and optimize it as well as the NDCG. Finally we create the first hierarchical landmarks retrieval dataset. We use a semi-automatic pipeline to create hierarchical labels, extending the large scale Google Landmarks v2 dataset. The hierarchical dataset is publicly available at https://github.com/cvdfoundation/google-landmark. Code will be released at https://github.com/elias-ramzi/SupRank., Comment: arXiv admin note: text overlap with arXiv:2207.04873

Strongly-shaped diverted negative triangularity (NT) plasmas in the DIII-D tokamak demonstrate simultaneous access to high normalized current, pressure, density, and confinement. NT plasmas are shown to exist across an expansive parameter space compatible with high fusion power production, revealing surprisingly good core stability properties that compare favorably to conventional positive triangularity plasmas in DIII-D. Non-dimensionalizing the operating space, edge safety factors below 3, normalized betas above 3, Greenwald density fractions above 1, and high-confinement mode (H-mode) confinement qualities above 1 are simultaneously observed, all with a robustly stable edge free from deleterious edge-localized mode instabilities. Scaling of the confinement time with engineering parameters reveals at least a linear dependence on plasma current although with significant power degradation, both in excess of expected H-mode scalings. These results increase confidence that NT plasmas are a viable approach to realize fusion power and open directions for future detailed study.

Vision-language foundation models such as CLIP have shown impressive zero-shot performance on many tasks and datasets, especially thanks to their free-text inputs. However, they struggle to handle some downstream tasks, such as fine-grained attribute detection and localization. In this paper, we propose a multitask fine-tuning strategy based on a positive/negative prompt formulation to further leverage the capacities of the vision-language foundation models. Using the CLIP architecture as baseline, we show strong improvements on bird fine-grained attribute detection and localization tasks, while also increasing the classification performance on the CUB200-2011 dataset. We provide source code for reproducibility purposes: it is available at https://github.com/FactoDeepLearning/MultitaskVLFM.

Recently, diffusion-based generative models have achieved remarkable success for image generation and edition. However, existing diffusion-based video editing approaches lack the ability to offer precise control over generated content that maintains temporal consistency in long-term videos. On the other hand, atlas-based methods provide strong temporal consistency but are costly to edit a video and lack spatial control. In this work, we introduce VidEdit, a novel method for zero-shot text-based video editing that guarantees robust temporal and spatial consistency. In particular, we combine an atlas-based video representation with a pre-trained text-to-image diffusion model to provide a training-free and efficient video editing method, which by design fulfills temporal smoothness. To grant precise user control over generated content, we utilize conditional information extracted from off-the-shelf panoptic segmenters and edge detectors which guides the diffusion sampling process. This method ensures a fine spatial control on targeted regions while strictly preserving the structure of the original video. Our quantitative and qualitative experiments show that VidEdit outperforms state-of-the-art methods on DAVIS dataset, regarding semantic faithfulness, image preservation, and temporal consistency metrics. With this framework, processing a single video only takes approximately one minute, and it can generate multiple compatible edits based on a unique text prompt. Project web-page at https://videdit.github.io, Comment: TMLR 2024. Project web-page at https://videdit.github.io

Out-of-distribution (OOD) detection is a critical requirement for the deployment of deep neural networks. This paper introduces the HEAT model, a new post-hoc OOD detection method estimating the density of in-distribution (ID) samples using hybrid energy-based models (EBM) in the feature space of a pre-trained backbone. HEAT complements prior density estimators of the ID density, e.g. parametric models like the Gaussian Mixture Model (GMM), to provide an accurate yet robust density estimation. A second contribution is to leverage the EBM framework to provide a unified density estimation and to compose several energy terms. Extensive experiments demonstrate the significance of the two contributions. HEAT sets new state-of-the-art OOD detection results on the CIFAR-10 / CIFAR-100 benchmark as well as on the large-scale Imagenet benchmark. The code is available at: https://github.com/MarcLafon/heatood.

In a series of high performance diverted discharges on DIII-D, we demonstrate that strong negative triangularity (NT) shaping robustly suppresses all edge-localized mode (ELM) activity over a wide range of plasma conditions: $\langle n\rangle=0.1-1.5\times10^{20}$m$^{-3}$, $P_\mathrm{aux}=0-15$MW and $|B_\mathrm{t}|=1-2.2$T, corresponding to $P_\mathrm{loss}/P_\mathrm{LH08}\sim8$. The full dataset is consistent with the theoretical prediction that magnetic shear in the NT edge inhibits access to ELMing H-mode regimes; all experimental pressure profiles are found to be at or below the infinite-$n$ ballooning stability limit. Importantly, we also report enhanced edge pressure gradients at strong NT that are significantly steeper than in traditional ELM-free L-mode plasmas and provide significant promise for NT reactor integration., Comment: 5 pages, 5 figures

The STEP (Stability, Transport, Equilibrium, and Pedestal) integrated-modeling tool has been developed in OMFIT to predict stable, tokamak equilibria self-consistently with core-transport and pedestal calculations. STEP couples theory-based codes to integrate a variety of physics, including MHD stability, transport, equilibrium, pedestal formation, and current-drive, heating, and fueling. The input/output of each code is interfaced with a centralized ITER-IMAS data structure, allowing codes to be run in any order and enabling open-loop, feedback, and optimization workflows. This paradigm simplifies the integration of new codes, making STEP highly extensible. STEP has been verified against a published benchmark of six different integrated models. Core-pedestal calculations with STEP have been successfully validated against individual DIII-D H-mode discharges and across more than 500 discharges of the $H_{98,y2}$ database, with a mean error in confinement time from experiment less than 19%. STEP has also reproduced results in less conventional DIII-D scenarios, including negative-central-shear and negative-triangularity plasmas. Predictive STEP modeling has been used to assess performance in several tokamak reactors. Simulations of a high-field, large-aspect-ratio reactor show significantly lower fusion power than predicted by a zero-dimensional study, demonstrating the limitations of scaling-law extrapolations. STEP predictions have found promising EXCITE scenarios, including a high-pressure, 80%-bootstrap-fraction plasma. ITER modeling with STEP has shown that pellet fueling enhances fusion gain in both the baseline and advanced-inductive scenarios. Finally, STEP predictions for the SPARC baseline scenario are in good agreement with published results from the physics basis., Comment: 15 pages, 11 figures Associated with invited talk at 63nd Annual Meeting of the APS Division of Plasma Physics: https://meetings.aps.org/Meeting/DPP21/Session/NI02.1 . The following article has been submitted to Physics of Plasmas. After it is published, it will be found at https://publishing.aip.org/resources/librarians/products/journals/

Estimating fluid dynamics is classically done through the simulation and integration of numerical models solving the Navier-Stokes equations, which is computationally complex and time-consuming even on high-end hardware. This is a notoriously hard problem to solve, which has recently been addressed with machine learning, in particular graph neural networks (GNN) and variants trained and evaluated on datasets of static objects in static scenes with fixed geometry. We attempt to go beyond existing work in complexity and introduce a new model, method and benchmark. We propose EAGLE, a large-scale dataset of 1.1 million 2D meshes resulting from simulations of unsteady fluid dynamics caused by a moving flow source interacting with nonlinear scene structure, comprised of 600 different scenes of three different types. To perform future forecasting of pressure and velocity on the challenging EAGLE dataset, we introduce a new mesh transformer. It leverages node clustering, graph pooling and global attention to learn long-range dependencies between spatially distant data points without needing a large number of iterations, as existing GNN methods do. We show that our transformer outperforms state-of-the-art performance on, both, existing synthetic and real datasets and on EAGLE. Finally, we highlight that our approach learns to attend to airflow, integrating complex information in a single iteration., Comment: Published as a conference paper at ICLR 2023

Predicting multiple trajectories for road users is important for automated driving systems: ego-vehicle motion planning indeed requires a clear view of the possible motions of the surrounding agents. However, the generative models used for multiple-trajectory forecasting suffer from a lack of diversity in their proposals. To avoid this form of collapse, we propose a novel method for structured prediction of diverse trajectories. To this end, we complement an underlying pretrained generative model with a diversity component, based on a determinantal point process (DPP). We balance and structure this diversity with the inclusion of knowledge-based quality constraints, independent from the underlying generative model. We combine these two novel components with a gating operation, ensuring that the predictions are both diverse and within the drivable area. We demonstrate on the nuScenes driving dataset the relevance of our compound approach, which yields significant improvements in the diversity and the quality of the generated trajectories.

Transformers have proved to be very effective for visual recognition tasks. In particular, vision transformers construct compressed global representations through self-attention and learnable class tokens. Multi-resolution transformers have shown recent successes in semantic segmentation but can only capture local interactions in high-resolution feature maps. This paper extends the notion of global tokens to build GLobal Attention Multi-resolution (GLAM) transformers. GLAM is a generic module that can be integrated into most existing transformer backbones. GLAM includes learnable global tokens, which unlike previous methods can model interactions between all image regions, and extracts powerful representations during training. Extensive experiments show that GLAM-Swin or GLAM-Swin-UNet exhibit substantially better performances than their vanilla counterparts on ADE20K and Cityscapes. Moreover, GLAM can be used to segment large 3D medical images, and GLAM-nnFormer achieves new state-of-the-art performance on the BCV dataset., Comment: Winter Conference on Applications of Computer Vision (WACV 2023)

Vision-Language Pretraining (VLP) and Foundation models have been the go-to recipe for achieving SoTA performance on general benchmarks. However, leveraging these powerful techniques for more complex vision-language tasks, such as cooking applications, with more structured input data, is still little investigated. In this work, we propose to leverage these techniques for structured-text based computational cuisine tasks. Our strategy, dubbed VLPCook, first transforms existing image-text pairs to image and structured-text pairs. This allows to pretrain our VLPCook model using VLP objectives adapted to the strutured data of the resulting datasets, then finetuning it on downstream computational cooking tasks. During finetuning, we also enrich the visual encoder, leveraging pretrained foundation models (e.g. CLIP) to provide local and global textual context. VLPCook outperforms current SoTA by a significant margin (+3.3 Recall@1 absolute improvement) on the task of Cross-Modal Food Retrieval on the large Recipe1M dataset. We conduct further experiments on VLP to validate their importance, especially on the Recipe1M+ dataset. Finally, we validate the generalization of the approach to other tasks (i.e, Food Recognition) and domains with structured text such as the Medical domain on the ROCO dataset. The code is available here: https://github.com/mshukor/VLPCook, Comment: Code: https://github.com/mshukor/VLPCook

Transformer models achieve state-of-the-art results for image segmentation. However, achieving long-range attention, necessary to capture global context, with high-resolution 3D images is a fundamental challenge. This paper introduces the Full resolutIoN mEmory (FINE) transformer to overcome this issue. The core idea behind FINE is to learn memory tokens to indirectly model full range interactions while scaling well in both memory and computational costs. FINE introduces memory tokens at two levels: the first one allows full interaction between voxels within local image regions (patches), the second one allows full interactions between all regions of the 3D volume. Combined, they allow full attention over high resolution images, e.g. 512 x 512 x 256 voxels and above. Experiments on the BCV image segmentation dataset shows better performances than state-of-the-art CNN and transformer baselines, highlighting the superiority of our full attention mechanism compared to recent transformer baselines, e.g. CoTr, and nnFormer.