Search

In quantised autoencoders, images are usually split into local patches, each encoded by one token. This representation is redundant in the sense that the same number of tokens is spend per region, regardless of the visual information content in that region. Adaptive discretisation schemes like quadtrees are applied to allocate tokens for patches with varying sizes, but this just varies the region of influence for a token which nevertheless remains a local descriptor. Modern architectures add an attention mechanism to the autoencoder which infuses some degree of global information into the local tokens. Despite the global context, tokens are still associated with a local image region. In contrast, our method is inspired by spectral decompositions which transform an input signal into a superposition of global frequencies. Taking the data-driven perspective, we learn custom basis functions corresponding to the codebook entries in our VQ-VAE setup. Furthermore, a decoder combines these basis functions in a non-linear fashion, going beyond the simple linear superposition of spectral decompositions. We can achieve this global description with an efficient transpose operation between features and channels and demonstrate our performance on compression.

Thermoelectric materials convert a temperature gradient into a voltage. This phenomenon is relatively well understood for inorganic materials, but much less so for organic semiconductors (OSs). These materials present a challenge because the strong thermal fluctuations of electronic coupling between the molecules result in partially delocalized charge carriers that cannot be treated with traditional theories for thermoelectricity. Here we develop a novel quantum dynamical simulation approach revealing in atomistic detail how the charge carrier wavefunction moves along a temperature gradient in an organic molecular crystal. We find that the wavefunction propagates from hot to cold in agreement with experiment and we obtain a Seebeck coefficient in good agreement with values obtained from experimental measurements that are also reported in this work. Detailed analysis of the dynamics reveals that the directional charge carrier motion is due to the gradient in thermal electronic disorder, more specifically in the spatial gradient of thermal fluctuations of electronic couplings. It causes an increase in the density of thermally accessible electronic states, the delocalization of states and the non-adiabatic coupling between states with decreasing temperature. As a result, the carrier wavefunction transitions with higher probability to a neighbouring electronic state towards the cold side compared to the hot side generating a thermoelectric current. Our dynamical perspective of thermoelectricity suggests that the temperature dependence of electronic disorder plays an important role in determining the magnitude of the Seebeck coefficient in this class of materials, opening new avenues for design of OSs with improved Seebeck coefficients.

This work investigates the in-context learning abilities of pretrained large language models (LLMs) when instructed to translate text from a low-resource language into a high-resource language as part of an automated machine translation pipeline. We conduct a set of experiments translating Southern Quechua to Spanish and examine the informativity of various types of context retrieved from a constrained database of digitized pedagogical materials (dictionaries and grammar lessons) and parallel corpora. Using both automatic and human evaluation of model output, we conduct ablation studies that manipulate (1) context type (morpheme translations, grammar descriptions, and corpus examples), (2) retrieval methods (automated vs. manual), and (3) model type. Our results suggest that even relatively small LLMs are capable of utilizing prompt context for zero-shot low-resource translation when provided a minimally sufficient amount of relevant linguistic information. However, the variable effects of context type, retrieval method, model type, and language-specific factors highlight the limitations of using even the best LLMs as translation systems for the majority of the world's 7,000+ languages and their speakers., Comment: Accepted at WMT24

The differential cross section for $\gamma p\rightarrow K^+\Lambda(1520)$ was measured from threshold to a centre-of-mass energy of 2090\,MeV at forward angles at the BGOOD experiment. The high statistical precision and resolution in centre-of-mass energy and angle allows a detailed characterisation of this low-momentum transfer kinematic region. The data agree with a previous LEPS measurement and support effective Lagrangian models that indicate that the contact term dominates the cross section near threshold., Comment: 6 pages, 7 figures

Recent advancements in generative models have sparked significant interest within the machine learning community. Particularly, diffusion models have demonstrated remarkable capabilities in synthesizing images and speech. Studies such as those by Lee et al. [19], Black et al. [4], Wang et al. [36], and Fan et al. [8] illustrate that Reinforcement Learning with Human Feedback (RLHF) can enhance diffusion models for image synthesis. However, due to architectural differences between these models and those employed in speech synthesis, it remains uncertain whether RLHF could similarly benefit speech synthesis models. In this paper, we explore the practical application of RLHF to diffusion-based text-to-speech synthesis, leveraging the mean opinion score (MOS) as predicted by UTokyo-SaruLab MOS prediction system [29] as a proxy loss. We introduce diffusion model loss-guided RL policy optimization (DLPO) and compare it against other RLHF approaches, employing the NISQA speech quality and naturalness assessment model [21] and human preference experiments for further evaluation. Our results show that RLHF can enhance diffusion-based text-to-speech synthesis models, and, moreover, DLPO can better improve diffusion models in generating natural and high quality speech audios.

The coherent reaction, $\gamma d \rightarrow \pi^0\eta d$ was studied with the BGOOD experiment at ELSA from threshold to a centre-of-mass energy of 3200\,MeV. A full kinematic reconstruction was made, with final state deuterons identified in the forward spectrometer and $\pi^0$ and $\eta$ decays in the central BGO Rugby Ball. The strength of the differential cross section exceeds what can be described by models of coherent photoproduction at forward angles by orders of magnitude. The distribution of the differential cross section has an excellent agreement with a model including quasi-free $\Delta \pi$ photoproduction, pion re-scattering and $N(1535)$ formation and subsequent nucleon coalescence to the deuteron. This also gives a reasonable description of the two-body invariant mass distributions and naturally explains the similar magnitudes of this channel and $\pi^0\pi^0 d$ coherent photoproduction., Comment: 7 pages, 5 figures

Seam carving is an image editing method that enable content-aware resizing, including operations like removing objects. However, the seam-finding strategy based on dynamic programming or graph-cut limits its applications to broader visual data formats and degrees of freedom for editing. Our observation is that describing the editing and retargeting of images more generally by a displacement field yields a generalisation of content-aware deformations. We propose to learn a deformation with a neural network that keeps the output plausible while trying to deform it only in places with low information content. This technique applies to different kinds of visual data, including images, 3D scenes given as neural radiance fields, or even polygon meshes. Experiments conducted on different visual data show that our method achieves better content-aware retargeting compared to previous methods., Comment: ECCV 2024

Franka Emika robots have gained significant popularity in research and education due to their exceptional versatility and advanced capabilities. This work introduces panda-py - a Python interface and framework designed to empower Franka Emika robotics with accessible and efficient programming. The panda-py interface enhances the usability of Franka Emika robots, enabling researchers and educators to interact with them more effectively. By leveraging Python's simplicity and readability, users can quickly grasp the necessary programming concepts for robot control and manipulation. Moreover, integrating panda-py with other widely used Python packages in domains such as computer vision and machine learning amplifies the robot's capabilities. Researchers can seamlessly leverage the vast ecosystem of Python libraries, thereby enabling advanced perception, decision-making, and control functionalities. This compatibility facilitates the efficient development of sophisticated robotic applications, integrating state-of-the-art techniques from diverse domains without the added complexity of ROS., Comment: 5 pages, 1 figure, 1 table, 7 listings

This paper explores how Generative Adversarial Networks (GANs) learn representations of phonological phenomena. We analyze how GANs encode contrastive and non-contrastive nasality in French and English vowels by applying the ciwGAN architecture (Begus 2021a). Begus claims that ciwGAN encodes linguistically meaningful representations with categorical variables in its latent space and manipulating the latent variables shows an almost one to one corresponding control of the phonological features in ciwGAN's generated outputs. However, our results show an interactive effect of latent variables on the features in the generated outputs, which suggests the learned representations in neural networks are different from the phonological representations proposed by linguists. On the other hand, ciwGAN is able to distinguish contrastive and noncontrastive features in English and French by encoding them differently. Comparing the performance of GANs learning from different languages results in a better understanding of what language specific features contribute to developing language specific phonological representations. We also discuss the role of training data frequencies in phonological feature learning.

Maltese is often described as having a hybrid morphological system resulting from extensive contact between Semitic and Romance language varieties. Such a designation reflects an etymological divide as much as it does a larger tradition in the literature to consider concatenative and non-concatenative morphological patterns as distinct in the language architecture. Using a combination of computational modeling and information theoretic methods, we quantify the extent to which the phonology and etymology of a Maltese singular noun may predict the morphological process (affixal vs. templatic) as well as the specific plural allomorph (affix or template) relating a singular noun to its associated plural form(s) in the lexicon. The results indicate phonological pressures shape the organization of the Maltese lexicon with predictive power that extends beyond that of a word's etymology, in line with analogical theories of language change in contact., Comment: Presented at the Annual Meeting of the Society for Computation in Linguistics 2023 (SCiL 2023)

Neural Radiance Fields (NeRFs) learn to represent a 3D scene from just a set of registered images. Increasing sizes of a scene demands more complex functions, typically represented by neural networks, to capture all details. Training and inference then involves querying the neural network millions of times per image, which becomes impractically slow. Since such complex functions can be replaced by multiple simpler functions to improve speed, we show that a hierarchy of Voronoi diagrams is a suitable choice to partition the scene. By equipping each Voronoi cell with its own NeRF, our approach is able to quickly learn a scene representation. We propose an intuitive partitioning of the space that increases quality gains during training by distributing information evenly among the networks and avoids artifacts through a top-down adaptive refinement. Our framework is agnostic to the underlying NeRF method and easy to implement, which allows it to be applied to various NeRF variants for improved learning and rendering speeds.

There is compelling evidence that charge carriers in organic semiconductors (OSs) self-localize in nano-scale space because of dynamic disorder. Yet, some OSs, in particular recently emerged high-mobility organic molecular crystals, feature reduced mobility at increasing temperature, a hallmark for delocalized band transport. Here we present the temperature-dependent mobility in two record-mobility OSs: DNTT (dinaphtho[2,3-b:2',3'-f]thieno[3,2-b]-thiophene), and its alkylated derivative, C8-DNTT-C8. By combining terahertz photoconductivity measurements with fully atomistic non-adiabatic molecular dynamics simulations, we show that while both crystals display a power-law decrease of the mobility (\mu) with temperature (T, following: \mu \propto T^(-n)), the exponent n differs substantially. Modelling provides n values in good agreement with experiments and reveals that the differences in the falloff parameter between the two chemically closely related semiconductors can be traced to the delocalization of the different states thermally accessible by charge carriers, which in turn depends on the specific electronic band structure of the two systems. The emerging picture is that of holes surfing on a dynamic manifold of vibrationally-dressed extended states with a temperature-dependent mobility that provides a sensitive fingerprint for the underlying density of states., Comment: 28 pages, 5 Figures

The Crystal Barrel is an electromagnetic calorimeter consisting of 1380 CsI(Tl) scintillators, and is currently installed at the CBELSA/TAPS experiment where it is used to detect decay products from photoproduction of mesons. The readout of the Crystal Barrel has been upgraded in order to integrate the detector into the first level of the trigger and to increase its sensitivity for neutral final states. The new readout uses avalanche photodiodes in the front-end and a dual back-end with branches optimized for energy and time measurement, respectively. An FPGA-based cluster finder processes the whole hit pattern within less than 100 ns. The important downside of APDs -- the temperature dependence of their gain -- is handled with a temperature stabilization and a compensating bias voltage supply. Additionally, a light pulser system allows the APDs' gains to be measured during beamtimes.

The convection velocity ($U_C$) of turbulent structures has been studied in adverse-pressure-gradient (APG) turbulent boundary layers (TBLs) for a wide range of Reynolds numbers $Re_\tau = 1400 - 4000$. The study is based on estimation of the convection velocity using decomposed streamwise skewness factor introduced in (Drozdz A., Elsner W., Int. J. of Heat and Fluid Flow. 63 (2017) 67\-74) and verified by means of two-point correlation method. It was shown that in the overlapping region of APG flows, the convection velocity profiles (when scaled in viscous units) reassemble the universal logarithmic law characteristic for the ZPG flows up to Clauser-Rota pressure gradient parameter $\beta \lesssim 19$ for the considered range of Reynolds number, what means that in the inner region of TBL the friction velocity in APG is not proportional to $U$ (as in ZPG) but to $U_C$ instead. The physical mechanism that explains the impact of increased convection velocity on the mean flow is proposed. The difference between $U_C$ and $U$ increases as a function of APG, which causes the stronger sweeping that enhances momentum transfer to the wall and compensates the weaker mean shear profile that is created by lower vorticity near the wall. This effect is a result of an enhancement of amplitude modulation of the small scales by large scale motion. The process becomes more pronounced as eddy density grows, so with increasing Re. The proposed model addresses a number of literature observations found in adverse pressure gradient flows which have been so far left without a well-founded explanation., Comment: 25 pages, 11 figures

This paper proposes a perceptual visual analysis regarding the personality of virtual humans. Many studies have presented findings regarding the way human beings perceive virtual humans with respect to their faces, body animation, motion in the virtual environment and etc. We are interested in investigating the way people perceive visual manifestations of virtual humans' personality traits when they are interactive and organized in groups. Many applications in games and movies can benefit from the findings regarding the perceptual analysis with the main goal to provide more realistic characters and improve the users' experience. We provide experiments with subjects and obtained results indicate that, although is very subtle, people perceive more the extraversion (the personality trait that we measured), into the crowds of virtual humans, when interacting with virtual humans behaviors, than when just observing as a spectator camera., Comment: 9 pages, 7 figures, 1 table

We study the effect of magnification in the Dark Energy Survey Year 3 analysis of galaxy clustering and galaxy-galaxy lensing, using two different lens samples: a sample of Luminous red galaxies, redMaGiC, and a sample with a redshift-dependent magnitude limit, MagLim. We account for the effect of magnification on both the flux and size selection of galaxies, accounting for systematic effects using the Balrog image simulations. We estimate the impact of magnification on the galaxy clustering and galaxy-galaxy lensing cosmology analysis, finding it to be a significant systematic for the MagLim sample. We show cosmological constraints from the galaxy clustering auto-correlation and galaxy-galaxy lensing signal with different magnifications priors, finding broad consistency in cosmological parameters in $\Lambda$CDM and $w$CDM. However, when magnification bias amplitude is allowed to be free, we find the two-point correlations functions prefer a different amplitude to the fiducial input derived from the image simulations. We validate the magnification analysis by comparing the cross-clustering between lens bins with the prediction from the baseline analysis, which uses only the auto-correlation of the lens bins, indicating systematics other than magnification may be the cause of the discrepancy. We show adding the cross-clustering between lens redshift bins to the fit significantly improves the constraints on lens magnification parameters and allows uninformative priors to be used on magnification coefficients, without any loss of constraining power or prior volume concerns., Comment: Version accepted for publication in MNRAS. 21 pages, 13 figures, See this https://www.darkenergysurvey.org/des-year-3-cosmology-results-papers/ URL for the full DES Y3 cosmology release

We constrain extensions to the $\Lambda$CDM model using measurements from the Dark Energy Survey's first three years of observations and external data. The DES data are the two-point correlation functions of weak gravitational lensing, galaxy clustering, and their cross-correlation. We use simulated data and blind analyses of real data to validate the robustness of our results. In many cases, constraining power is limited by the absence of nonlinear predictions that are reliable at our required precision. The models are: dark energy with a time-dependent equation of state, non-zero spatial curvature, sterile neutrinos, modifications of gravitational physics, and a binned $\sigma_8(z)$ model which serves as a probe of structure growth. For the time-varying dark energy equation of state evaluated at the pivot redshift we find $(w_{\rm p}, w_a)= (-0.99^{+0.28}_{-0.17},-0.9\pm 1.2)$ at 68% confidence with $z_{\rm p}=0.24$ from the DES measurements alone, and $(w_{\rm p}, w_a)= (-1.03^{+0.04}_{-0.03},-0.4^{+0.4}_{-0.3})$ with $z_{\rm p}=0.21$ for the combination of all data considered. Curvature constraints of $\Omega_k=0.0009\pm 0.0017$ and effective relativistic species $N_{\rm eff}=3.10^{+0.15}_{-0.16}$ are dominated by external data. For massive sterile neutrinos, we improve the upper bound on the mass $m_{\rm eff}$ by a factor of three compared to previous analyses, giving 95% limits of $(\Delta N_{\rm eff},m_{\rm eff})\leq (0.28, 0.20\, {\rm eV})$. We also constrain changes to the lensing and Poisson equations controlled by functions $\Sigma(k,z) = \Sigma_0 \Omega_{\Lambda}(z)/\Omega_{\Lambda,0}$ and $\mu(k,z)=\mu_0 \Omega_{\Lambda}(z)/\Omega_{\Lambda,0}$ respectively to $\Sigma_0=0.6^{+0.4}_{-0.5}$ from DES alone and $(\Sigma_0,\mu_0)=(0.04\pm 0.05,0.08^{+0.21}_{-0.19})$ for the combination of all data. Overall, we find no significant evidence for physics beyond $\Lambda$CDM., Comment: Updated to match published version and fix a citation reference. 46 pages, 25 figures, data available at https://dev.des.ncsa.illinois.edu/releases/y3a2/Y3key-extensions

Measurements of target asymmetries and double-polarization observables for the reaction $\gamma p\to p\pi^0\pi^0$ are reported. The data were taken with the CBELSA/TAPS experiment at the ELSA facility (Bonn University) using the Bonn frozen-spin butanol (C$_4$H$_9$OH) target, which provided transversely polarized protons. Linearly polarized photons were produced via bremsstrahlung off a diamond crystal. The data cover the photon energy range from $E_{\gamma}$=650 MeV to $E_{\gamma}$=2600 MeV and nearly the complete angular range. The results have been included in the BnGa partial wave analysis. Experimental results and the fit agree very well. Observed systematic differences in the branching ratios for decays of $N^*$ and $\Delta^*$ resonances are attributed to the internal structure of these excited nucleon states. Resonances which can be assigned to SU(6)$\times$O(3) two-oscillator configurations show larger branching ratios to intermediate states with non-zero intrinsic orbital angular momenta than resonances assigned to one-oscillator configurations., Comment: 21 pages, 27 figures

We present cosmological constraints from the analysis of two-point correlation functions between galaxy positions and galaxy lensing measured in Dark Energy Survey (DES) Year 3 data and measurements of cosmic microwave background (CMB) lensing from the South Pole Telescope (SPT) and Planck. When jointly analyzing the DES-only two-point functions and the DES cross-correlations with SPT+Planck CMB lensing, we find $\Omega_{\rm m} = 0.344\pm 0.030$ and $S_8 \equiv \sigma_8 (\Omega_{\rm m}/0.3)^{0.5} = 0.773\pm 0.016$, assuming $\Lambda$CDM. When additionally combining with measurements of the CMB lensing autospectrum, we find $\Omega_{\rm m} = 0.306^{+0.018}_{-0.021}$ and $S_8 = 0.792\pm 0.012$. The high signal-to-noise of the CMB lensing cross-correlations enables several powerful consistency tests of these results, including comparisons with constraints derived from cross-correlations only, and comparisons designed to test the robustness of the galaxy lensing and clustering measurements from DES. Applying these tests to our measurements, we find no evidence of significant biases in the baseline cosmological constraints from the DES-only analyses or from the joint analyses with CMB lensing cross-correlations. However, the CMB lensing cross-correlations suggest possible problems with the correlation function measurements using alternative lens galaxy samples, in particular the redMaGiC galaxies and high-redshift MagLim galaxies, consistent with the findings of previous studies. We use the CMB lensing cross-correlations to identify directions for further investigating these problems., Comment: 20 pages, 15 figures

We use the small scales of the Dark Energy Survey (DES) Year-3 cosmic shear measurements, which are excluded from the DES Year-3 cosmological analysis, to constrain the baryonic feedback. To model the baryonic feedback, we adopt a baryonic correction model and use the numerical package \texttt{Baccoemu} to accelerate the evaluation of the baryonic nonlinear matter power spectrum. We design our analysis pipeline to focus on the constraints of the baryonic suppression effects, utilizing the implication given by a principal component analysis on the Fisher forecasts. Our constraint on the baryonic effects can then be used to better model and ameliorate the effects of baryons in producing cosmological constraints from the next generation large-scale structure surveys. We detect the baryonic suppression on the cosmic shear measurements with a $\sim 2 \sigma$ significance. The characteristic halo mass for which half of the gas is ejected by baryonic feedback is constrained to be $M_c > 10^{13.2} h^{-1} M_{\odot}$ (95\% C.L.). The best-fit baryonic suppression is $\sim 5\%$ at $k=1.0 {\rm Mpc}\ h^{-1}$ and $\sim 15\%$ at $k=5.0 {\rm Mpc} \ h^{-1}$. Our findings are robust with respect to the assumptions about the cosmological parameters, specifics of the baryonic model, and intrinsic alignments., Comment: 20 pages, 10 figures. DES Collaboration, Year-3 analysis

Humans can make predictions on various time scales and hierarchical levels. Thereby, the learning of event encodings seems to play a crucial role. In this work we model the development of hierarchical predictions via autonomously learned latent event codes. We present a hierarchical recurrent neural network architecture, whose inductive learning biases foster the development of sparsely changing latent state that compress sensorimotor sequences. A higher level network learns to predict the situations in which the latent states tend to change. Using a simulated robotic manipulator, we demonstrate that the system (i) learns latent states that accurately reflect the event structure of the data, (ii) develops meaningful temporal abstract predictions on the higher level, and (iii) generates goal-anticipatory behavior similar to gaze behavior found in eye-tracking studies with infants. The architecture offers a step towards the autonomous learning of compressed hierarchical encodings of gathered experiences and the exploitation of these encodings to generate adaptive behavior., Comment: accepted at ICDL 2022