Search

Current neural network models of primate vision focus on replicating overall levels of behavioral accuracy, often neglecting perceptual decisions' rich, dynamic nature. Here, we introduce a novel computational framework to model the dynamics of human behavioral choices by learning to align the temporal dynamics of a recurrent neural network (RNN) to human reaction times (RTs). We describe an approximation that allows us to constrain the number of time steps an RNN takes to solve a task with human RTs. The approach is extensively evaluated against various psychophysics experiments. We also show that the approximation can be used to optimize an "ideal-observer" RNN model to achieve an optimal tradeoff between speed and accuracy without human data. The resulting model is found to account well for human RT data. Finally, we use the approximation to train a deep learning implementation of the popular Wong-Wang decision-making model. The model is integrated with a convolutional neural network (CNN) model of visual processing and evaluated using both artificial and natural image stimuli. Overall, we present a novel framework that helps align current vision models with human behavior, bringing us closer to an integrated model of human vision., Comment: Published at NeurIPS 2024

Large Language Model (LLM) fine tuning is underutilized in the field of medicine. Two of the most common methods of fine tuning are Supervised Fine Tuning (SFT) and Direct Preference Optimization (DPO), but there is little guidance informing users when to use either technique. In this investigation, we compare the performance of SFT and DPO for five common natural language tasks in medicine: Classification with text data, Classification with numeric data, Clinical Reasoning, Summarization, and Clinical Triage. We find that SFT alone is sufficient for Classification with text data, whereas DPO improves performance for the more complex tasks of Clinical Reasoning, Summarization and Clinical Triage. Our results establish the role and importance of DPO fine tuning within medicine, and consequently call attention to current software gaps that prevent widespread deployment of this technique.

We study the design of robust and agile controllers for hybrid underactuated systems. Our approach breaks down the task of creating a stabilizing controller into: 1) learning a mapping that is invariant under optimal control, and 2) driving the actuated coordinates to the output of that mapping. This approach, termed Zero Dynamics Policies, exploits the structure of underactuation by restricting the inputs of the target mapping to the subset of degrees of freedom that cannot be directly actuated, thereby achieving significant dimension reduction. Furthermore, we retain the stability and constraint satisfaction of optimal control while reducing the online computational overhead. We prove that controllers of this type stabilize hybrid underactuated systems and experimentally validate our approach on the 3D hopping platform, ARCHER. Over the course of 3000 hops the proposed framework demonstrates robust agility, maintaining stable hopping while rejecting disturbances on rough terrain., Comment: 8 pages, 6 figures, IROS '24

Stabilizing underactuated systems is an inherently challenging control task due to fundamental limitations on how the control input affects the unactuated dynamics. Decomposing the system into actuated (output) and unactuated (zero) coordinates provides useful insight as to how input enters the system dynamics. In this work, we leverage the structure of this decomposition to formalize the idea of Zero Dynamics Policies (ZDPs) -- a mapping from the unactuated coordinates to desired actuated coordinates. Specifically, we show that a ZDP exists in a neighborhood of the origin, and prove that combining output stabilization with a ZDP results in stability of the full system state. We detail a constructive method of obtaining ZDPs in a neighborhood of the origin, and propose a learning-based approach which leverages optimal control to obtain ZDPs with much larger regions of attraction. We demonstrate that such a paradigm can be used to stabilize the canonical underactuated system of the cartpole, and showcase an improvement over the nominal performance of LQR., Comment: 8 pages, 2 figures, CDC 2024

There is a continuing interest in using standard language constructs for accelerated computing in order to avoid (sometimes vendor-specific) external APIs. For Fortran codes, the {\tt do concurrent} (DC) loop has been successfully demonstrated on the NVIDIA platform. However, support for DC on other platforms has taken longer to implement. Recently, Intel has added DC GPU offload support to its compiler, as has HPE for AMD GPUs. In this paper, we explore the current portability of using DC across GPU vendors using the in-production solar surface flux evolution code, HipFT. We discuss implementation and compilation details, including when/where using directive APIs for data movement is needed/desired compared to using a unified memory system. The performance achieved on both data center and consumer platforms is shown., Comment: 9 pages, 7 figures

In a hierarchical, dark matter-dominated Universe, stellar mass functions (SMFs), galaxy merger rates, star formation histories (SFHs), satellite abundances, and intracluster light, should all be intimately connected observables. However, the systematics affecting observations still prevent universal and uniform measurements of, for example, the SMF and the SFHs, inevitably preventing theoretical models to compare with multiple data sets robustly and simultaneously. We here present our holistic semi-empirical model DECODE (Discrete statistical sEmi-empiriCal mODEl) that converts via abundance matching dark matter merger trees into galaxy assembly histories, using different SMFs in input and predicting all other observables in output in a fully data-driven and self-consistent fashion with minimal assumptions. We find that: 1) weakly evolving or nearly constant SMFs below the knee ($M_\star \lesssim 10^{11} \, M_\odot$) are the best suited to generate star formation histories aligned with those inferred from MaNGA, SDSS, GAMA, and, more recently, JWST; 2) the evolution of satellites after infall only affects the satellite abundances and star formation histories of massive central galaxies but not their merger histories; 3) the resulting SFR-$M_\star$ relation is lower in normalization by a factor of $\sim 2$ with respect to observations, with a flattening at high masses more pronounced in the presence of mergers; 4) the latest data on intracluster light can be reproduced if mass loss from mergers is included in the models. Our findings are pivotal in acting as pathfinder to test the self-consistency of the high-quality data from, e.g., JWST and Euclid., Comment: 21 pages, 12 figures, MNRAS accepted

Environmental effects are believed to play an important yet poorly understood role in triggering accretion events onto the supermassive black holes (SMBHs) of galaxies (Active Galactic Nuclei; AGN). Massive clusters, which represent the densest structures in the Universe, provide an excellent laboratory to isolate environmental effects and study their impact on black hole growth. In this work, we critically review observational evidence for the preferential activation of SMBHs in the outskirts of galaxy clusters. We develop a semi-empirical model under the assumption that the incidence of AGN in galaxies is independent of environment. We demonstrate that the model is broadly consistent with recent observations on the AGN halo occupation at $z=0.2$, although it may overpredict satellite AGN in massive halos at that low redshift. We then use this model to interpret the projected radial distribution of X-ray sources around high redshift ($z\approx1$) massive ($>5 \times 10^{14} \, M_\odot$) clusters, which show excess counts outside their virial radius. Such an excess naturally arises in our model as a result of sample variance. Up to 20% of the simulated projected radial distributions show excess counts similar to the observations, which are however, because of background/foreground AGN and hence, not physically associated with the cluster. Our analysis emphasises the importance of projection effects and shows that current observations of $z\approx1$ clusters remain inconclusive on the activation of SMBHs during infall., Comment: Submitted to MNRAS, 15 pages, 12 Figures

Evolutionary models suggest that the initial growth phases of active galactic nuclei (AGN) are dust-enshrouded, and characterised by jet/wind outflows that should gradually clear the interstellar medium (ISM) in the host by heating and/or expelling the surrounding gas. eFEDSJ091157.4$+$014327 (z$\sim$0.6) was selected from X-ray samples for its characteristics that are similar to sources with outflows which include red, obscured and X-ray luminous. We aim to explore the environment around this red quasar and characterise kinematics within the system. We used spatially resolved spectroscopic data from Multi Unit Spectroscopic Explorer (MUSE) with an average seeing of 0.6" to construct flux, velocity and velocity dispersions maps. We found that the quasar is embedded in an interacting and merging system with three other galaxies $\sim$ 50 kpc from its nucleus. Spatially resolved kinematics reveal that the quasar has extended ionised outflows of up to 9.2 kpc with positive and negative velocities up to 1000 km s$^{-1}$ and -1200 km s$^{-1}$, respectively. The velocity dispersion (W$_{80}$) ranges from 600-1800 km s$^{-1}$. We associate the presence of turbulent and high-velocity components with the outflow. The total mass outflow rate is estimated to be $\sim$ 10 M$_{\odot}$ yr$^{-1}$ and kinetic power of 2x10$^{42}$ erg s$^{-1}$. The kinetic coupling efficiencies range from 0.01%-0.03% and the momentum boosts are $\sim$ 0.2. These low values indicate that the ionised outflow is not energetically relevant. These values don't align with the theoretical predictions of both radiation-pressure-driven outflows and energy-conserving mechanisms. However, note that our results are based only on the ionised phase while theoretical predictions are multiphase. Moreover, the mass loading factor of $\sim$ 5 is an indication that these outflows are more likely AGN-driven than star formation-driven., Comment: Version after addressing referee comments, accepted in Astronomy and Astrophysics(A&A), 17 pages, 12 figures, 1 table

Perception systems for ornithopters face severe challenges. The harsh vibrations and abrupt movements caused during flapping are prone to produce motion blur and strong lighting condition changes. Their strict restrictions in weight, size, and energy consumption also limit the type and number of sensors to mount onboard. Lightweight traditional cameras have become a standard off-the-shelf solution in many flapping-wing designs. However, bioinspired event cameras are a promising solution for ornithopter perception due to their microsecond temporal resolution, high dynamic range, and low power consumption. This paper presents an experimental comparison between frame-based and an event-based camera. Both technologies are analyzed considering the particular flapping-wing robot specifications and also experimentally analyzing the performance of well-known vision algorithms with data recorded onboard a flapping-wing robot. Our results suggest event cameras as the most suitable sensors for ornithopters. Nevertheless, they also evidence the open challenges for event-based vision on board flapping-wing robots.

In this paper, we combine the techniques of enlargement of filtrations and stochastic control theory to establish an extension of the verification theorem, where the coefficients of the stochastic controlled equation are adapted to the underlying filtration and the controls are adapted to a bigger filtration $\mathbf{G}$ than the one generated by the corresponding Brownian motion $B$. Using the forward integral defined by Russo and Vallois \cite{RV1993}, we show that there is a $\mathbf{G}$-adapted optimal control with respect to a certain cost functional if and only if the Brownian motion $B$ is a $\mathbf{G}$-semimartingale. The extended verification theorem allows us to study a financial market with an insider in order to take advantage of the extra information that the insider has from the beginning. Finally, we consider two examples throughout the extended verification theorem. These problems appear in financial markets with an insider., Comment: 16 pages, new version including funding information

One of the most impactful findings in computational neuroscience over the past decade is that the object recognition accuracy of deep neural networks (DNNs) correlates with their ability to predict neural responses to natural images in the inferotemporal (IT) cortex. This discovery supported the long-held theory that object recognition is a core objective of the visual cortex, and suggested that more accurate DNNs would serve as better models of IT neuron responses to images. Since then, deep learning has undergone a revolution of scale: billion parameter-scale DNNs trained on billions of images are rivaling or outperforming humans at visual tasks including object recognition. Have today's DNNs become more accurate at predicting IT neuron responses to images as they have grown more accurate at object recognition? Surprisingly, across three independent experiments, we find this is not the case. DNNs have become progressively worse models of IT as their accuracy has increased on ImageNet. To understand why DNNs experience this trade-off and evaluate if they are still an appropriate paradigm for modeling the visual system, we turn to recordings of IT that capture spatially resolved maps of neuronal activity elicited by natural images. These neuronal activity maps reveal that DNNs trained on ImageNet learn to rely on different visual features than those encoded by IT and that this problem worsens as their accuracy increases. We successfully resolved this issue with the neural harmonizer, a plug-and-play training routine for DNNs that aligns their learned representations with humans. Our results suggest that harmonized DNNs break the trade-off between ImageNet accuracy and neural prediction accuracy that assails current DNNs and offer a path to more accurate models of biological vision.

The current trend in developing machine learning models for reading comprehension and logical reasoning tasks is focused on improving the models' abilities to understand and utilize logical rules. This work focuses on providing a novel loss function and accompanying model architecture that has more interpretable components than some other models by representing a common strategy employed by humans when given reading comprehension and logical reasoning tasks. Our strategy involves emphasizing relative accuracy over absolute accuracy and can theoretically produce the correct answer with incomplete knowledge. We examine the effectiveness of this strategy to solve reading comprehension and logical reasoning questions. The models were evaluated on the ReClor dataset, a challenging reading comprehension and logical reasoning benchmark. We propose the polytuplet loss function, which forces prioritization of learning the relative correctness of answer choices over learning the true accuracy of each choice. Our results indicate that models employing polytuplet loss outperform existing baseline models, though further research is required to quantify the benefits it may present., Comment: Accepted to FICC 2023, Revised to correct clerical errors

This paper explores the role of small-scale environment ($<1$ Mpc) in modulating accretion events onto supermassive black holes by studying the incidence of Active Galactic Nuclei (AGN) in massive clusters of galaxies. A flexible, data-driven semi-empirical model is developed based on a minimal set of parameters and under the zero order assumption that the incidence of AGN in galaxies is independent of environment. This is used to predict how the fraction of X-ray selected AGN among galaxies in massive dark matter halos ($\gtrsim 3\times 10^{14}\,M_{\odot}$) evolves with redshift and reveal tensions with observations. At high redshift, $z\sim1.2$, the model underpredicts AGN fractions, particularly at high X-ray luminosities, $L_X(\rm 2-10\,keV) \gtrsim 10^{44}\, erg \, s^{-1}$. At low redshift, $z\sim0.2$, the model estimates fractions of moderate luminosity AGN ($L_X(\rm 2-10\,keV) \gtrsim 10^{43}\, erg \, s^{-1}$) that are a factor of $2-3$ higher than the observations. These findings reject the zero order assumption on which the semi-empirical model hinges and point to a strong and redshift-dependent influence of the small-scale environment on the growth of black holes. Cluster of galaxies appear to promote AGN activity relative to the model expectation at $z\sim1.2$ and suppress it close to the present day. These trends could be explained by the increasing gas content of galaxies toward higher redshift combined with an efficient triggering of AGN at earlier times in galaxies that fall onto clusters., Comment: Accepted in Monthly Notices of the Royal Astronomical Society

Forward invariance is a long-studied property in control theory that is used to certify that a dynamical system stays within some pre-specified set of states for all time, and also admits robustness guarantees (e.g., the certificate holds under perturbations). We propose a general framework for training and provably certifying robust forward invariance in Neural ODEs. We apply this framework to provide certified safety in robust continuous control. To our knowledge, this is the first instance of training Neural ODE policies with such non-vacuous certified guarantees. In addition, we explore the generality of our framework by using it to certify adversarial robustness for image classification.

Vision transformers are nowadays the de-facto choice for image classification tasks. There are two broad categories of classification tasks, fine-grained and coarse-grained. In fine-grained classification, the necessity is to discover subtle differences due to the high level of similarity between sub-classes. Such distinctions are often lost as we downscale the image to save the memory and computational cost associated with vision transformers (ViT). In this work, we present an in-depth analysis and describe the critical components for developing a system for the fine-grained categorization of plants from herbarium sheets. Our extensive experimental analysis indicated the need for a better augmentation technique and the ability of modern-day neural networks to handle higher dimensional images. We also introduce a convolutional transformer architecture called Conviformer which, unlike the popular Vision Transformer (ConViT), can handle higher resolution images without exploding memory and computational cost. We also introduce a novel, improved pre-processing technique called PreSizer to resize images better while preserving their original aspect ratios, which proved essential for classifying natural plants. With our simple yet effective approach, we achieved SoTA on Herbarium 202x and iNaturalist 2019 dataset., Comment: 12 pages; 4 Figures; 8 Tables

The weight, space, and power limitations of small aerial vehicles often prevent the application of modern control techniques without significant model simplifications. Moreover, high-speed agile behavior, such as that exhibited in drone racing, make these simplified models too unreliable for safety-critical control. In this work, we introduce the concept of time-varying backup controllers (TBCs): user-specified maneuvers combined with backup controllers that generate reference trajectories which guarantee the safety of nonlinear systems. TBCs reduce conservatism when compared to traditional backup controllers and can be directly applied to multi-agent coordination to guarantee safety. Theoretically, we provide conditions under which TBCs strictly reduce conservatism, describe how to switch between several TBC's and show how to embed TBCs in a multi-agent setting. Experimentally, we verify that TBCs safely increase operational freedom when filtering a pilot's actions and demonstrate robustness and computational efficiency when applied to decentralized safety filtering of two quadrotors., Comment: Submitted to IROS 2022

This work presents Neural Gaits, a method for learning dynamic walking gaits through the enforcement of set invariance that can be refined episodically using experimental data from the robot. We frame walking as a set invariance problem enforceable via control barrier functions (CBFs) defined on the reduced-order dynamics quantifying the underactuated component of the robot: the zero dynamics. Our approach contains two learning modules: one for learning a policy that satisfies the CBF condition, and another for learning a residual dynamics model to refine imperfections of the nominal model. Importantly, learning only over the zero dynamics significantly reduces the dimensionality of the learning problem while using CBFs allows us to still make guarantees for the full-order system. The method is demonstrated experimentally on an underactuated bipedal robot, where we are able to show agile and dynamic locomotion, even with partially unknown dynamics.

With the increasing prevalence of complex vision-based sensing methods for use in obstacle identification and state estimation, characterizing environment-dependent measurement errors has become a difficult and essential part of modern robotics. This paper presents a self-supervised learning approach to safety-critical control. In particular, the uncertainty associated with stereo vision is estimated, and adapted online to new visual environments, wherein this estimate is leveraged in a safety-critical controller in a robust fashion. To this end, we propose an algorithm that exploits the structure of stereo-vision to learn an uncertainty estimate without the need for ground-truth data. We then robustify existing Control Barrier Function-based controllers to provide safety in the presence of this uncertainty estimate. We demonstrate the efficacy of our method on a quadrupedal robot in a variety of environments. When not using our method safety is violated. With offline training alone we observe the robot is safe, but overly-conservative. With our online method the quadruped remains safe and conservatism is reduced., Comment: 7 pages, 4 figures, conference publication at ICRA 2022

We propose a method for training ordinary differential equations by using a control-theoretic Lyapunov condition for stability. Our approach, called LyaNet, is based on a novel Lyapunov loss formulation that encourages the inference dynamics to converge quickly to the correct prediction. Theoretically, we show that minimizing Lyapunov loss guarantees exponential convergence to the correct solution and enables a novel robustness guarantee. We also provide practical algorithms, including one that avoids the cost of backpropagating through a solver or using the adjoint method. Relative to standard Neural ODE training, we empirically find that LyaNet can offer improved prediction performance, faster convergence of inference dynamics, and improved adversarial robustness. Our code available at https://github.com/ivandariojr/LyapunovLearning .

Following the trend of other safety-critical industries like automotive and avionics, the space domain is witnessing an increase in the on-board computing performance demands. This raise in performance needs comes from both control and payload parts of the spacecraft and calls for advanced electronics systems able to provide high computational power under the constraints of the harsh space environment. On the non-technical side, for strategic reasons it is mandatory to get European independence on the used computing technology. In this project, we study the applicability of embedded GPUs in space, which have shown a dramatic improvement of their performance per-watt ratio coming from their proliferation in consumer markets based on competitive European technology. To that end, we perform an analysis of the existing space application domains to identify which software domains can benefit from their use. Moreover, we survey the embedded GPU domain in order to assess whether embedded GPUs can provide the required computational power and identify the challenges which need to be addressed for their adoption in space. In this paper, we describe the steps followed in the project, as well as a summary of results obtained from our analyses so far in the project., Comment: Elsevier Journal of Microprocessors and Microsystems, September 2020

The Institute of Materials and Processes, IMP, of the University of Applied Sciences in Karlsruhe, Germany in cooperation with VDI Verein Deutscher Ingenieure e.V, AEN Automotive Engineering Network and their cooperation partners present their competences of AI-based solution approaches in the production engineering field. The online congress KI 4 Industry on November 12 and 13, 2020, showed what opportunities the use of artificial intelligence offers for medium-sized manufacturing companies, SMEs, and where potential fields of application lie. The main purpose of KI 4 Industry is to increase the transfer of knowledge, research and technology from universities to small and medium-sized enterprises, to demystify the term AI and to encourage companies to use AI-based solutions in their own value chain or in their products., Comment: Editors: Matthias Feiner and Manuel Schoellhorn, 72 pages, 48 figures, in German, Conference proceedings KI 4 Industry, 79 pages in total

Recently Bonet and Geffner have shown that first-order representations for planning domains can be learned from the structure of the state space without any prior knowledge about the action schemas or domain predicates. For this, the learning problem is formulated as the search for a simplest first-order domain description D that along with information about instances I_i (number of objects and initial state) determine state space graphs G(P_i) that match the observed state graphs G_i where P_i = (D, I_i). The search is cast and solved approximately by means of a SAT solver that is called over a large family of propositional theories that differ just in the parameters encoding the possible number of action schemas and domain predicates, their arities, and the number of objects. In this work, we push the limits of these learners by moving to an answer set programming (ASP) encoding using the CLINGO system. The new encodings are more transparent and concise, extending the range of possible models while facilitating their exploration. We show that the domains introduced by Bonet and Geffner can be solved more efficiently in the new approach, often optimally, and furthermore, that the approach can be easily extended to handle partial information about the state graphs as well as noise that prevents some states from being distinguished.

We consider the problem of reaching a propositional goal condition in fully-observable non-deterministic (FOND) planning under a general class of fairness assumptions that are given explicitly. The fairness assumptions are of the form A/B and say that state trajectories that contain infinite occurrences of an action a from A in a state s and finite occurrence of actions from B, must also contain infinite occurrences of action a in s followed by each one of its possible outcomes. The infinite trajectories that violate this condition are deemed as unfair, and the solutions are policies for which all the fair trajectories reach a goal state. We show that strong and strong-cyclic FOND planning, as well as QNP planning, a planning model introduced recently for generalized planning, are all special cases of FOND planning with fairness assumptions of this form which can also be combined. FOND+ planning, as this form of planning is called, combines the syntax of FOND planning with some of the versatility of LTL for expressing fairness constraints. A new planner is implemented by reducing FOND+ planning to answer set programs, and the performance of the planner is evaluated in comparison with FOND and QNP planners, and LTL synthesis tools., Comment: Extended version of ICAPS-21 paper

We present a straightforward and efficient way to control unstable robotic systems using an estimated dynamics model. Specifically, we show how to exploit the differentiability of Gaussian Processes to create a state-dependent linearized approximation of the true continuous dynamics that can be integrated with model predictive control. Our approach is compatible with most Gaussian process approaches for system identification, and can learn an accurate model using modest amounts of training data. We validate our approach by learning the dynamics of an unstable system such as a segway with a 7-D state space and 2-D input space (using only one minute of data), and we show that the resulting controller is robust to unmodelled dynamics and disturbances, while state-of-the-art control methods based on nominal models can fail under small perturbations. Code is open sourced at https://github.com/learning-and-control/core .

Decision trees are ubiquitous in machine learning for their ease of use and interpretability. Yet, these models are not typically employed in reinforcement learning as they cannot be updated online via stochastic gradient descent. We overcome this limitation by allowing for a gradient update over the entire tree that improves sample complexity affords interpretable policy extraction. First, we include theoretical motivation on the need for policy-gradient learning by examining the properties of gradient descent over differentiable decision trees. Second, we demonstrate that our approach equals or outperforms a neural network on all domains and can learn discrete decision trees online with average rewards up to 7x higher than a batch-trained decision tree. Third, we conduct a user study to quantify the interpretability of a decision tree, rule list, and a neural network with statistically significant results ($p < 0.001$).