Papers made digestable
Our architecture simplifies the obstacle-perception
problem to that of place-dependent change detection. While we use the method with VT&R, it
can be generalized to suit arbitrary path-following applications.
Visual Teach and Repeat 3 (VT&R3), a generalization of stereo VT&R, achieves
long-term autonomous path-following using topometric mapping and localization
from a single rich sensor stream. In this paper, we improve the capabilities of
a LiDAR implementation of VT&R3 to reliably detect and avoid obstacles in
changing environments. Our architecture simplifies the obstacle-perception
problem to that of place-dependent change detection. We then extend the
behaviour of generic sample-based motion planners to better suit the
teach-and-repeat problem structure by introducing a new edge-cost metric paired
with a curvilinear planning space. The resulting planner generates naturally
smooth paths that avoid local obstacles while minimizing lateral path deviation
to best exploit prior terrain knowledge. While we use the method with VT&R, it
can be generalized to suit arbitrary path-following applications. Experimental
results from online run-time analysis, unit testing, and qualitative
experiments on a differential drive robot show the promise of the technique for
reliable long-term autonomous operation in complex unstructured environments.
Authors: Jordy Sehn, Yuchen Wu, Timothy D. Barfoot.
The statistical and design considerations that pertain to
dose optimization are discussed. The sample size savings range from 16.6% to 27.3%,
depending on the design and scenario, with a mean savings of 22.1%.
The traditional more-is-better dose selection paradigm, developed based on
cytotoxic chemotherapeutics, is often problematic When applied to the
development of novel molecularly targeted agents (e.g., kinase inhibitors,
monoclonal antibodies, and antibody-drug conjugates). The US Food and Drug
Administration (FDA) initiated Project Optimus to reform the dose optimization
and dose selection paradigm in oncology drug development and call for more
attention to benefit-risk consideration.
We systematically investigated the operating characteristics of the seamless
phase 2-3 design as a strategy for dose optimization, where in stage 1
(corresponding to phase 2) patients are randomized to multiple doses, with or
without a control; and in stage 2 (corresponding to phase 3) the efficacy of
the selected optimal dose is evaluated with a randomized concurrent control or
historical control. Depending on whether the concurrent control is included and
the type of endpoints used in stages 1 and 2, we describe four types of
seamless phase 2-3 dose-optimization designs, which are suitable for different
clinical settings. The statistical and design considerations that pertain to
dose optimization are discussed. Simulation shows that dose optimization phase
2-3 designs are able to control the familywise type I error rates and yield
appropriate statistical power with substantially smaller sample size than the
conventional approach. The sample size savings range from 16.6% to 27.3%,
depending on the design and scenario, with a mean savings of 22.1%. Due to the
interim dose selection, the phase 2-3 dose-optimization design is logistically
and operationally more challenging, and should be carefully planned and
implemented to ensure trial integrity.
Authors: Liyun Jiang, Ying Yuan.
We significantly improve performance using properties of the posterior
in our active learning scheme and for the definition of the GP prior. In
particular we account for the expected dynamical range of the posterior in
different dimensionalities. We test our model against a number of synthetic and
cosmological examples.
We present the GPry algorithm for fast Bayesian inference of general
(non-Gaussian) posteriors with a moderate number of parameters. GPry does not
need any pre-training, special hardware such as GPUs, and is intended as a
drop-in replacement for traditional Monte Carlo methods for Bayesian inference.
Our algorithm is based on generating a Gaussian Process surrogate model of the
log-posterior, aided by a Support Vector Machine classifier that excludes
extreme or non-finite values. An active learning scheme allows us to reduce the
number of required posterior evaluations by two orders of magnitude compared to
traditional Monte Carlo inference. Our algorithm allows for parallel
evaluations of the posterior at optimal locations, further reducing wall-clock
times. We significantly improve performance using properties of the posterior
in our active learning scheme and for the definition of the GP prior. In
particular we account for the expected dynamical range of the posterior in
different dimensionalities. We test our model against a number of synthetic and
cosmological examples. GPry outperforms traditional Monte Carlo methods when
the evaluation time of the likelihood (or the calculation of theoretical
observables) is of the order of seconds; for evaluation times of over a minute
it can perform inference in days that would take months using traditional
methods. GPry is distributed as an open source Python package (pip install
gpry) and can also be found at https://github.com/jonaselgammal/GPry.
Authors: Jonas El Gammal, Nils Schöneberg, Jesús Torrado, Christian Fidler.
We consider the fundamental scheduling problem of minimizing the sum of
weighted completion times on a single machine in the non-clairvoyant setting. However, to the best of our knowledge, this concept has never been considered
for the total completion time objective in the non-clairvoyant model. This implies
a performance guarantee of $(1+3\sqrt{3})\approx 6.197$ for the deterministic
algorithm and of $\approx 3.032$ for the randomized version.
We consider the fundamental scheduling problem of minimizing the sum of
weighted completion times on a single machine in the non-clairvoyant setting.
While no non-preemptive algorithm is constant competitive, Motwani, Phillips,
and Torng (SODA '93) proved that the simple preemptive round robin procedure is
$2$-competitive and that no better competitive ratio is possible, initiating a
long line of research focused on preemptive algorithms for generalized variants
of the problem. As an alternative model, Shmoys, Wein, and Williamson (FOCS
'91) introduced kill-and-restart schedules, where running jobs may be killed
and restarted from scratch later, and analyzed then for the makespan objective.
However, to the best of our knowledge, this concept has never been considered
for the total completion time objective in the non-clairvoyant model.
We contribute to both models: First we give for any $b > 1$ a tight analysis
for the natural $b$-scaling kill-and-restart strategy for scheduling jobs
without release dates, as well as for a randomized variant of it. This implies
a performance guarantee of $(1+3\sqrt{3})\approx 6.197$ for the deterministic
algorithm and of $\approx 3.032$ for the randomized version. Second, we show
that the preemptive Weighted Shortest Elapsed Time First (WSETF) rule is
$2$-competitive for jobs released in an online fashion over time, matching the
lower bound by Motwani et al. Using this result as well as the competitiveness
of round robin for multiple machines, we prove performance guarantees of
adaptions of the $b$-scaling algorithm to online release dates and unweighted
jobs on identical parallel machines.
Authors: Sven Jäger, Guillaume Sagnol, Daniel Schmidt genannt Waldschmidt, Philipp Warode.
Frozen pretrained models have become a viable alternative to the
pretraining-then-finetuning paradigm for transfer learning. With this work, we hope to
bring greater attention to this promising path of freezing pretrained image
models.
Frozen pretrained models have become a viable alternative to the
pretraining-then-finetuning paradigm for transfer learning. However, with
frozen models there are relatively few parameters available for adapting to
downstream tasks, which is problematic in computer vision where tasks vary
significantly in input/output format and the type of information that is of
value. In this paper, we present a study of frozen pretrained models when
applied to diverse and representative computer vision tasks, including object
detection, semantic segmentation and video action recognition. From this
empirical analysis, our work answers the questions of what pretraining task
fits best with this frozen setting, how to make the frozen setting more
flexible to various downstream tasks, and the effect of larger model sizes. We
additionally examine the upper bound of performance using a giant frozen
pretrained model with 3 billion parameters (SwinV2-G) and find that it reaches
competitive performance on a varied set of major benchmarks with only one
shared frozen base network: 60.0 box mAP and 52.2 mask mAP on COCO object
detection test-dev, 57.6 val mIoU on ADE20K semantic segmentation, and 81.7
top-1 accuracy on Kinetics-400 action recognition. With this work, we hope to
bring greater attention to this promising path of freezing pretrained image
models.
Authors: Yutong Lin, Ze Liu, Zheng Zhang, Han Hu, Nanning Zheng, Stephen Lin, Yue Cao.
While our formulas are generic, we discuss
explicit examples of phenomenological relevance considering the physics case of
the curvaton field. We carefully assess under which conditions the conventional
perturbative approach can be trusted.
We develop an exact formalism for the computation of the abundance of
primordial black holes (PBHs) in the presence of local non-gaussianity (NG) in
the curvature perturbation field. For the first time, we include NG going
beyond the widely used quadratic and cubic approximations, and consider a
completely generic functional form. Adopting threshold statistics of the
compaction function, we address the computation of the abundance both for
narrow and broad power spectra. While our formulas are generic, we discuss
explicit examples of phenomenological relevance considering the physics case of
the curvaton field. We carefully assess under which conditions the conventional
perturbative approach can be trusted. In the case of a narrow power spectrum,
this happens only if the perturbative expansion is pushed beyond the quadratic
order (with the optimal order of truncation that depends on the width of the
spectrum). Most importantly, we demonstrate that the perturbative approach is
intrinsically flawed when considering broad spectra, in which case only the
non-perturbative computation captures the correct result. Finally, we describe
the phenomenological relevance of our results for the connection between the
abundance of PBHs and the stochastic gravitational wave (GW) background related
to their formation. As NGs modify the amplitude of perturbations necessary to
produce a given PBHs abundance and boost PBHs production at large scales for
broad spectra, modelling these effects is crucial to connect the PBH scenario
to its signatures at current and future GWs experiments.
Authors: Giacomo Ferrante, Gabriele Franciolini, Antonio Junior Iovino, Alfredo Urbano.
Vector autoregressions (VARs) are popular in analyzing economic time series. However, VARs can be over-parameterized if the numbers of variables and lags are moderately large. Firstly, we determine the rank by imposing the Multiplicative Gamma Prior to margins, i.e. elements in the decomposition, and accelerate the computation with an adaptive inferential scheme.
Vector autoregressions (VARs) are popular in analyzing economic time series.
However, VARs can be over-parameterized if the numbers of variables and lags
are moderately large. Tensor VAR, a recent solution to overparameterization,
treats the coefficient matrix as a third-order tensor and estimates the
corresponding tensor decomposition to achieve parsimony. In this paper, the
inference of Tensor VARs is inspired by the literature on factor models.
Firstly, we determine the rank by imposing the Multiplicative Gamma Prior to
margins, i.e. elements in the decomposition, and accelerate the computation
with an adaptive inferential scheme. Secondly, to obtain interpretable margins,
we propose an interweaving algorithm to improve the mixing of margins and
introduce a post-processing procedure to solve column permutations and
sign-switching issues. In the application of the US macroeconomic data, our
models outperform standard VARs in point and density forecasting and yield
interpretable results consistent with the US economic history.
Authors: Yiyong Luo, Jim E. Griffin.
Measurement is a fundamental enabler of network applications such as load
balancing, attack detection and mitigation, and traffic engineering. Modern approaches attain a
constant runtime by removing small items in bulk and retaining the largest $q$
items at all times. Yet, these approaches are bottlenecked by an expensive
quantile calculation method. We
demonstrate the benefit of our approach by designing a novel weighted heavy
hitters data structure that is faster and more accurate than the existing
alternatives. Here, we combine our previous techniques with a lazy deletion of
small entries, which expiates the maintenance process and increases the
accuracy. We also demonstrate the applicability of our algorithmic approach in
a general algorithmic scope by implementing the LRFU cache policy with a
constant update time.
Measurement is a fundamental enabler of network applications such as load
balancing, attack detection and mitigation, and traffic engineering. A key
building block in many critical measurement tasks is \emph{q-MAX}, where we
wish to find the largest $q$ values in a number stream. A standard approach of
maintaining a heap of the largest $q$ values ordered results in logarithmic
runtime, which is too slow for large measurements. Modern approaches attain a
constant runtime by removing small items in bulk and retaining the largest $q$
items at all times. Yet, these approaches are bottlenecked by an expensive
quantile calculation method.
We propose SQUID, a method that redesigns q-MAX to allow the use of
\emph{approximate quantiles}, which we can compute efficiently, thereby
accelerating the solution and, subsequently, many measurement tasks. We
demonstrate the benefit of our approach by designing a novel weighted heavy
hitters data structure that is faster and more accurate than the existing
alternatives. Here, we combine our previous techniques with a lazy deletion of
small entries, which expiates the maintenance process and increases the
accuracy. We also demonstrate the applicability of our algorithmic approach in
a general algorithmic scope by implementing the LRFU cache policy with a
constant update time. Furthermore, we also show the practicality of SQUID for
improving real-world networked systems, by implementing a P4 prototype of SQUID
for in-network caching and demonstrating how SQUID enables a wide spectrum of
score-based caching policies directly on a P4 switch.
Authors: Ran Ben-Basat, Gil Einziger, Wenchen Han, Bilal Tayh.
In a reconfiguration problem, given a problem and two feasible solutions of the problem, the task is to find a sequence of transformations to reach from one solution to the other such that every intermediate state is also a feasible solution to the problem. In this paper, we study the distributed spanning tree reconfiguration problem and we define a new reconfiguration step, called $k$-simultaneous add and delete, in which every node is allowed to add at most $k$ edges and delete at most $k$ edges such that multiple nodes do not add or delete the same edge.
In a reconfiguration problem, given a problem and two feasible solutions of
the problem, the task is to find a sequence of transformations to reach from
one solution to the other such that every intermediate state is also a feasible
solution to the problem. In this paper, we study the distributed spanning tree
reconfiguration problem and we define a new reconfiguration step, called
$k$-simultaneous add and delete, in which every node is allowed to add at most
$k$ edges and delete at most $k$ edges such that multiple nodes do not add or
delete the same edge.
We first observe that, if the two input spanning trees are rooted, then we
can do the reconfiguration using a single $1$-simultaneous add and delete step
in one round in the CONGEST model. Therefore, we focus our attention towards
unrooted spanning trees and show that transforming an unrooted spanning tree
into another using a single $1$-simultaneous add and delete step requires
$\Omega(n)$ rounds in the LOCAL model. We additionally show that transforming
an unrooted spanning tree into another using a single $2$-simultaneous add and
delete step can be done in $O(\log n)$ rounds in the CONGEST model.
Authors: Siddharth Gupta, Manish Kumar, Shreyas Pai.
As we show, iterative inversion can converge to
the desired inverse mapping, but under rather strict conditions on the mapping
itself. We next apply iterative inversion to learn control. With a VQ-VAE embedding, and a transformer-based policy, we demonstrate
non-trivial continuous control on several tasks. We also report improved
performance on imitating diverse behaviors compared to reward based methods.
We formulate learning for control as an $\textit{inverse problem}$ --
inverting a dynamical system to give the actions which yield desired behavior.
The key challenge in this formulation is a $\textit{distribution shift}$ -- the
learning agent only observes the forward mapping (its actions' consequences) on
trajectories that it can execute, yet must learn the inverse mapping for
inputs-outputs that correspond to a different, desired behavior. We propose a
general recipe for inverse problems with a distribution shift that we term
$\textit{iterative inversion}$ -- learn the inverse mapping under the current
input distribution (policy), then use it on the desired output samples to
obtain new inputs, and repeat. As we show, iterative inversion can converge to
the desired inverse mapping, but under rather strict conditions on the mapping
itself.
We next apply iterative inversion to learn control. Our input is a set of
demonstrations of desired behavior, given as video embeddings of trajectories,
and our method iteratively learns to imitate trajectories generated by the
current policy, perturbed by random exploration noise. We find that constantly
adding the demonstrated trajectory embeddings $\textit{as input}$ to the policy
when generating trajectories to imitate, a-la iterative inversion, steers the
learning towards the desired trajectory distribution. To the best of our
knowledge, this is the first exploration of learning control from the viewpoint
of inverse problems, and our main advantage is simplicity -- we do not require
rewards, and only employ supervised learning, which easily scales to
state-of-the-art trajectory embedding techniques and policy representations.
With a VQ-VAE embedding, and a transformer-based policy, we demonstrate
non-trivial continuous control on several tasks. We also report improved
performance on imitating diverse behaviors compared to reward based methods.
Authors: Gal Leibovich, Guy Jacob, Or Avner, Gal Novik, Aviv Tamar.
This logic can express a wide variety of problems that escape the expressibility potential of FOL. Using the same technique we prove that model-checking for FOL+SDP can be done in quadratic time on classes of graphs with bounded Euler genus.
The disjoint paths logic, FOL+DP, is an extension of First-Order Logic (FOL)
with the extra atomic predicate ${\sf dp}_k(x_1,y_1,\ldots,x_k,y_k),$
expressing the existence of internally vertex-disjoint paths between $x_i$ and
$y_i,$ for $i\in\{1,\ldots, k\}$. This logic can express a wide variety of
problems that escape the expressibility potential of FOL. We prove that for
every proper minor-closed graph class, model-checking for FOL+DP can be done in
quadratic time. We also introduce an extension of FOL+DP, namely the scattered
disjoint paths logic, FOL+SDP, where we further consider the atomic predicate
$s{\sf -sdp}_k(x_1,y_1,\ldots,x_k,y_k),$ demanding that the disjoint paths are
within distance bigger than some fixed value $s$. Using the same technique we
prove that model-checking for FOL+SDP can be done in quadratic time on classes
of graphs with bounded Euler genus.
Authors: Petr A. Golovach, Giannos Stamoulis, Dimitrios M. Thilikos.
Our experiments show that the SNanoBERT
model is 25.9x smaller and 38.8x faster than SRoBERTa while achieving
comparable results on well-known benchmarks. Hence, making it suitable for
deploying with ASR models on edge devices.
Many studies have examined the shortcomings of word error rate (WER) as an
evaluation metric for automatic speech recognition (ASR) systems, particularly
when used for spoken language understanding tasks such as intent recognition
and dialogue systems. In this paper, we propose Hybrid-SD
($\text{H}_{\text{SD}}$), a new hybrid evaluation metric for ASR systems that
takes into account both semantic correctness and error rate. To generate
sentence dissimilarity scores (SD), we built a fast and lightweight SNanoBERT
model using distillation techniques. Our experiments show that the SNanoBERT
model is 25.9x smaller and 38.8x faster than SRoBERTa while achieving
comparable results on well-known benchmarks. Hence, making it suitable for
deploying with ASR models on edge devices. We also show that
$\text{H}_{\text{SD}}$ correlates more strongly with downstream tasks such as
intent recognition and named-entity recognition (NER).
Authors: Zitha Sasindran, Harsha Yelchuri, Supreeth Rao, T. V. Prabhakar.
Almost three-quarters of them could be labelled as long-term stable. We compared their analysis based on the AMD criterion with our results. The possible discrepancies are discussed.
Here we present an initial look at the dynamics and stability of 178
multiplanetary systems which are already confirmed and listed in the NASA
Exoplanet Archive. To distinguish between the chaotic and regular nature of a
system, the value of the MEGNO indicator for each system was determined. Almost
three-quarters of them could be labelled as long-term stable. Only 45 studied
systems show chaotic behaviour. We consequently investigated the effects of the
number of planets and their parameters on the system stability. A comparison of
results obtained using the MEGNO indicator and machine-learning algorithm SPOCK
suggests that the SPOCK could be used as an effective tool for reviewing the
stability of multiplanetary systems. A similar study was already published by
Laskar and Petit in 2017. We compared their analysis based on the AMD criterion
with our results. The possible discrepancies are discussed.
Authors: Pavol Gajdoš, Martin Vaňko.
Most of daily embedded
devices are real-time systems, e.g. airplanes, cars, trains, spatial probes,
etc. The time required by a program for its end-to-end execution is called its
response time.
Real-time systems are a set of programs, a scheduling policy and a system
architecture, constrained by timing requirements. Most of daily embedded
devices are real-time systems, e.g. airplanes, cars, trains, spatial probes,
etc. The time required by a program for its end-to-end execution is called its
response time. Usually, upper-bounds of response times are computed in order to
provide safe deadline miss probabilities. In this paper, we propose a suited
re-parametrization of the inverse Gaussian mixture distribution adapted to
response times of real-time systems and the estimation of deadline miss
probabilities. The parameters and their associated deadline miss probabilities
are estimated with an adapted Expectation-Maximization algorithm.
Authors: Kevin Zagalo, Olena Verbytska, Liliana Cucu-Grosjean, Avner Bar-Hen.
Cs will be routinely evaporated in the source by means of specific ovens. Monitoring the evaporation rate and the distribution of Cs inside the source is fundamental to get the desired performances on the ITER HNB. A Laser Absorption Spectroscopy diagnostic will be installed in SPIDER for a quantitative estimation of Cs density. From the absorption spectra the line-integrated density of Cs at ground state will be measured. The design of this diagnostic for SPIDER is presented, with details of the layout and of the key components.
The ITER Heating Neutral Beam (HNB) injector is required to deliver 16.7 MW
power into the plasma from a neutralised beam of H-/D- negative ions, produced
by an RF source and accelerated up to 1 MeV. To enhance the H-/D- production,
the surface of the acceleration system grid facing the source (the plasma grid)
will be coated with Cs because of its low work function. Cs will be routinely
evaporated in the source by means of specific ovens. Monitoring the evaporation
rate and the distribution of Cs inside the source is fundamental to get the
desired performances on the ITER HNB. In order to proper design the source of
the ITER HNB and to identify the best operation practices for it, the prototype
RF negative ion source SPIDER has been developed and built in the Neutral Beam
Test Facility at Consorzio RFX. A Laser Absorption Spectroscopy diagnostic will
be installed in SPIDER for a quantitative estimation of Cs density. By using a
wavelength tunable laser, the diagnostic will measure the absorption spectrum
of the 852 nm line along 4 lines of sight, parallel to the plasma grid surface
and close to it. From the absorption spectra the line-integrated density of Cs
at ground state will be measured. The design of this diagnostic for SPIDER is
presented, with details of the layout and of the key components. A preliminary
installation of the diagnostic on the test stand for Cs ovens is also
described, together with its first experimental results; the effect of ground
state depopulation on collected measurements is discussed and partially
corrected.
Authors: M. Barbisan, R. Pasqualotto, A. Rizzolo.