Search

Not all learnable parameters (e.g., weights) contribute equally to a neural network's decision function. In fact, entire layers' parameters can sometimes be reset to random values with little to no impact on the model's decisions. We revisit earlier studies that examined how architecture and task complexity influence this phenomenon and ask: is this phenomenon also affected by how we train the model? We conducted experimental evaluations on a diverse set of ImageNet-1k classification models to explore this, keeping the architecture and training data constant but varying the training pipeline. Our findings reveal that the training method strongly influences which layers become critical to the decision function for a given task. For example, improved training regimes and self-supervised training increase the importance of early layers while significantly under-utilizing deeper layers. In contrast, methods such as adversarial training display an opposite trend. Our preliminary results extend previous findings, offering a more nuanced understanding of the inner mechanics of neural networks. Code: https://github.com/paulgavrikov/layer_criticality, Comment: Accepted at the Interpretable AI: Past, Present and Future Workshop at NeurIPS 2024

Two-fluid plasma flow equations describe the flow of ions and electrons with different densities, velocities, and pressures. We consider the ideal plasma flow i.e. we ignore viscous, resistive, and collision effects. The resulting system of equations has flux consisting of three independent components, one for ions, one for electrons, and a linear Maxwell's equation flux for the electromagnetic fields. The coupling of these components is via source terms. In this article, we present {conservative} second-order finite difference schemes that ensure the consistent evolution of the divergence constraints on the electric and magnetic fields. The key idea is to design a numerical solver for Maxwell's equations using the multidimensional Riemann solver at the vertices, ensuring discrete divergence constraints; for the fluid parts, we use an entropy-stable discretization. The proposed schemes are co-located, second-order accurate, entropy stable, and ensure divergence-free evolution of the magnetic field. We use explicit and IMplicit-EXplicit (IMEX) schemes for time discretizations. To demonstrate the accuracy, stability, and divergence constraint-preserving ability of the proposed schemes, we present several test cases in one and two dimensions. We also compare the numerical results with those obtained from schemes with no divergence cleaning and those employing perfectly hyperbolic Maxwell (PHM) equations-based divergence cleaning methods for Maxwell's equations., Comment: Accepted in Journal of Scientific Computing

Reinforcement Learning with Human Feedback (RLHF) is at the core of fine-tuning methods for generative AI models for language and images. Such feedback is often sought as rank or preference feedback from human raters, as opposed to eliciting scores since the latter tends to be noisy. On the other hand, RL theory and algorithms predominantly assume that a reward feedback is available. In particular, approaches for online learning that can be helpful in adaptive data collection via active learning cannot incorporate offline preference data. In this paper, we adopt a finite-armed linear bandit model as a prototypical model of online learning. We consider an offline preference dataset to be available generated by an expert of unknown 'competence'. We propose $\texttt{warmPref-PS}$, a posterior sampling algorithm for online learning that can be warm-started with an offline dataset with noisy preference feedback. We show that by modeling the 'competence' of the expert that generated it, we are able to use such a dataset most effectively. We support our claims with novel theoretical analysis of its Bayesian regret, as well as, extensive empirical evaluation of an approximate loss function that optimizes for infinitely many arms, and performs substantially better ($25$ to $50\%$ regret reduction) than baselines.

In this paper, we present the $\texttt{e-COP}$ algorithm, the first policy optimization algorithm for constrained Reinforcement Learning (RL) in episodic (finite horizon) settings. Such formulations are applicable when there are separate sets of optimization criteria and constraints on a system's behavior. We approach this problem by first establishing a policy difference lemma for the episodic setting, which provides the theoretical foundation for the algorithm. Then, we propose to combine a set of established and novel solution ideas to yield the $\texttt{e-COP}$ algorithm that is easy to implement and numerically stable, and provide a theoretical guarantee on optimality under certain scaling assumptions. Through extensive empirical analysis using benchmarks in the Safety Gym suite, we show that our algorithm has similar or better performance than SoTA (non-episodic) algorithms adapted for the episodic setting. The scalability of the algorithm opens the door to its application in safety-constrained Reinforcement Learning from Human Feedback for Large Language or Diffusion Models.

Image restoration networks are usually comprised of an encoder and a decoder, responsible for aggregating image content from noisy, distorted data and to restore clean, undistorted images, respectively. Data aggregation as well as high-resolution image generation both usually come at the risk of involving aliases, i.e.~standard architectures put their ability to reconstruct the model input in jeopardy to reach high PSNR values on validation data. The price to be paid is low model robustness. In this work, we show that simply providing alias-free paths in state-of-the-art reconstruction transformers supports improved model robustness at low costs on the restoration performance. We do so by proposing BOA-Restormer, a transformer-based image restoration model that executes downsampling and upsampling operations partly in the frequency domain to ensure alias-free paths along the entire model while potentially preserving all relevant high-frequency information., Comment: Tags: Adversarial attack, image restoration, image deblurring, frequency sampling

Lossless image compression is required in various applications to reduce storage or transmission costs of images, while requiring the reconstructed images to have zero information loss compared to the original. Existing lossless image compression methods either have simple design but poor compression performance, or complex design, better performance, but with no performance guarantees. In our endeavor to develop a lossless image compression method with low complexity and guaranteed performance, we argue that compressibility of a color image is essentially derived from the patterns in its spatial structure, intensity variations, and color variations. Thus, we divide the overall design of a lossless image compression scheme into three parts that exploit corresponding redundancies. We further argue that the binarized version of an image captures its fundamental spatial structure. In this first part of our work, we propose a scheme for lossless compression of binary images. The proposed scheme first learns dictionaries of $16\times16$, $8\times8$, $4\times4$, and $2\times 2$ square pixel patterns from various datasets of binary images. It then uses these dictionaries to encode binary images. These dictionaries have various interesting properties that are further exploited to construct an efficient and scalable scheme. Our preliminary results show that the proposed scheme consistently outperforms existing conventional and learning based lossless compression approaches, and provides, on average, as much as $1.5\times$ better performance than a common general purpose lossless compression scheme (WebP), more than $3\times$ better performance than a state of the art learning based scheme, and better performance than a specialized scheme for binary image compression (JBIG2)., Comment: Corrected axes labels on Figs 8-11, slightly reorganized content with 13 pages, 11 figures, and 5 tables

We present a detailed investigation of photometric, spectroscopic, and polarimetric observations of the Type II SN 2023ixf. Earlier studies have provided compelling evidence for a delayed shock breakout from a confined dense circumstellar matter (CSM) enveloping the progenitor star. The temporal evolution of polarization in SN~2023ixf revealed three distinct peaks in polarization evolution at 1.4 d, 6.4 d, and 79.2 d, indicating an asymmetric dense CSM, an aspherical shock front and clumpiness in the low-density extended CSM, and an aspherical inner ejecta/He-core. SN 2023ixf displayed two dominant axes, one along the CSM-outer ejecta and the other along the inner ejecta/He-core, showcasing the independent origin of asymmetry in the early and late evolution. The argument for an aspherical shock front is further strengthened by the presence of a high-velocity broad absorption feature in the blue wing of the Balmer features in addition to the P-Cygni absorption post 16 d. Hydrodynamical light curve modeling indicated a progenitor of 10 solar mass with a radius of 470 solar radii and explosion energy of 2e51 erg, along with 0.06 solar mass of 56-Ni, though these properties are not unique due to modeling degeneracies. The modeling also indicated a two-zone CSM: a confined dense CSM extending up to 5e14 cm, with a mass-loss rate of 1e-2 solar mass per year, and an extended CSM spanning from 5e14 cm to at least 1e16cm with a mass-loss rate of 1e-4 solar mass per year, both assuming a wind-velocity of 10 km/s. The early nebular phase observations display an axisymmetric line profile of [OI], red-ward attenuation of the emission of Halpha post 125 days, and flattening in the Ks-band, marking the onset of dust formation., Comment: 32 pages, 15 figures, 1 Table, Accepted in the Astrophysical Journal

We give for the first time, theoretical estimates of ground-state-to-ground-state (GS-to-GS) electron-capture (EC) branch decay rates of $^{44}$Ti, $^{57}$Co, and $^{139}$Ce. The nuclear-structure calculations have been done exploiting the nuclear shell model (NSM) with well-established Hamiltonians and an advanced theory of $\beta$ decay. In the absence of experimental measurements of these GS-to-GS branches, these estimates are of utmost importance for terrestrial searches of axion-like particles (ALPs). Predictions are made for EC-decay rates of 2$^{nd}$-forbidden unique (FU) and 2$^{nd}$-forbidden non-unique (FNU) EC transitions that can potentially mimic nuclear axion production in experiments designed to detect ALPs in nuclear environments.

Let $F$ (over $\mathbb{Q}$) be a totally real number field of narrow class number $1$. We generalize a result of Kohnen on the determination of half integral weight modular forms by their Fourier coefficients supported on squarefree (algebraic) integers. We also give a soft proof that infinitely many Fourier coefficients supported on squarefree integers are non-vanishing., Comment: Comments and suggestions welcome

The uplink sum-throughput of distributed massive multiple-input-multiple-output (mMIMO) networks depends majorly on Access point (AP)-User Equipment (UE) association and power control. The AP-UE association and power control both are important problems in their own right in distributed mMIMO networks to improve scalability and reduce front-haul load of the network, and to enhance the system performance by mitigating the interference and boosting the desired signals, respectively. Unlike previous studies, which focused primarily on addressing these two problems separately, this work addresses the uplink sum-throughput maximization problem in distributed mMIMO networks by solving the joint AP-UE association and power control problem, while maintaining Quality-of-Service (QoS) requirements for each UE. To improve scalability, we present an l1-penalty function that delicately balances the trade-off between spectral efficiency (SE) and front-haul signaling load. Our proposed methodology leverages fractional programming, Lagrangian dual formation, and penalty functions to provide an elegant and effective iterative solution with guaranteed convergence. Extensive numerical simulations validate the efficacy of the proposed technique for maximizing sum-throughput while considering the joint AP-UE association and power control problem, demonstrating its superiority over approaches that address these problems individually. Furthermore, the results show that the introduced penalty function can help us effectively control the maximum front-haul load., Comment: This will be presented in the IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) 2024

This white paper introduces aoip.ai, a groundbreaking open-source SDK incorporating peer-to-peer technology and advanced AI integration to transform VoIP and IoT applications. It addresses key market challenges by enhancing data security, elevating communication quality, and providing greater flexibility for developers and users. Developed in collaboration with Carnegie Mellon University, aoip.ai sets a new standard for decentralized and democratized communication solutions.

Pixel-wise predictions are required in a wide variety of tasks such as image restoration, image segmentation, or disparity estimation. Common models involve several stages of data resampling, in which the resolution of feature maps is first reduced to aggregate information and then increased to generate a high-resolution output. Previous works have shown that resampling operations are subject to artifacts such as aliasing. During downsampling, aliases have been shown to compromise the prediction stability of image classifiers. During upsampling, they have been leveraged to detect generated content. Yet, the effect of aliases during upsampling has not yet been discussed w.r.t. the stability and robustness of pixel-wise predictions. While falling under the same term (aliasing), the challenges for correct upsampling in neural networks differ significantly from those during downsampling: when downsampling, some high frequencies can not be correctly represented and have to be removed to avoid aliases. However, when upsampling for pixel-wise predictions, we actually require the model to restore such high frequencies that can not be encoded in lower resolutions. The application of findings from signal processing is therefore a necessary but not a sufficient condition to achieve the desirable output. In contrast, we find that the availability of large spatial context during upsampling allows to provide stable, high-quality pixel-wise predictions, even when fully learning all filter weights., Comment: Accepted at ECCV 2024

Detecting anomalies in a daily time series with a weekly pattern is a common task with a wide range of applications. A typical way of performing the task is by using decomposition method. However, the method often generates false positive results where a data point falls within its weekly range but is just off from its weekday position. We refer to this type of anomalies as "in-season anomalies", and propose a k-parameter approach to address the issue. The approach provides configurable extra tolerance for in-season anomalies to suppress misleading alerts while preserving real positives. It yields favorable result., Comment: 5 pages, 7 figures

Within the ambit of VoIP (Voice over Internet Protocol) telecommunications, the complexities introduced by acoustic transformations merit rigorous analysis. This research, rooted in the exploration of proprietary sender-side denoising effects, meticulously evaluates platforms such as Google Meets and Zoom. The study draws upon the Deep Noise Suppression (DNS) 2020 dataset, ensuring a structured examination tailored to various denoising settings and receiver interfaces. A methodological novelty is introduced via Blinder-Oaxaca decomposition, traditionally an econometric tool, repurposed herein to analyze acoustic-phonetic perturbations within VoIP systems. To further ground the implications of these transformations, psychoacoustic metrics, specifically PESQ and STOI, were used to explain of perceptual quality and intelligibility. Cumulatively, the insights garnered underscore the intricate landscape of VoIP-influenced acoustic dynamics. In addition to the primary findings, a multitude of metrics are reported, extending the research purview. Moreover, out-of-domain benchmarking for both time and time-frequency domain speech enhancement models is included, thereby enhancing the depth and applicability of this inquiry.