  We study the evolution of the Vis-NIR reflectance spectrum of salty granular
ice analog samples in a simulation chamber under conditions relevant to the
surface of Europa. A novel application and custom calibration of a thermopile
sensor enabled the measurement of the surface temperature of the samples in far
infrared emission. This allows the kinetics of the observed changes to be
scaled to equivalent timescales on Europa. We observed significant changes in
the depth and shape of the broad water absorption bands for all samples on
timescales of a few thousand years of equatorial conditions on Europa. This
effect should be taken into account if quantitative predictions about bulk
composition are made based on remote-sensing data. A narrow absorption feature
attributed to hydrohalite formed during the sublimation of the sodium chloride
sample. We used near-infrared spectroscopy in an irradiation chamber to assess
the stability of this narrow feature under electron irradiation. We find that
the radiation environment present on Europa dehydrates the hydrohalite on short
timescales. Therefore, we expect hydrohalite not to be present on the surface,
unless erupted very recently (< 10 yr) or located in thermal anomalies (> 145
K). Thus, a detection of hydrohalite would clearly indicate recent activity.


Evolution of granular salty ice analogs for Europa: Sublimation and
  Irradiation

  The inclusion of spin effects in the binary dynamics for black hole and
neutron stars is crucial for the computation of gravitational wave observables.
Worldline supersymmetric models have shown to be particularly efficient at this
task up to quadratic order in spin, but progress at higher orders has been
hampered by no-go-theorems. In this work we propose a novel approach to
overcome this problem by extending the supersymmetry beyond minimal coupling.
We demonstrate the potential of this approach by computing an all-order in spin
and linear in curvature, manifestly supersymmetric Hamiltonian, as well as a
cubic order in spin Hamiltonian in arbitrary spacetime dimensions. In doing so,
we identify a criterion that uniquely determines the Kerr geometry in terms of
worldline supersymmetry. Equipped with these Hamiltonians, we demonstrate the
exponentiation of three-point and Compton amplitudes using the recently
proposed Generalized Wilson line approach.


Higher-spin effects in black hole and neutron star binary dynamics:
  worldline supersymmetry beyond minimal coupling

  Test-Time Scaling (TTS) refers to approaches that improve reasoning
performance by allocating extra computation during inference, without altering
the model's parameters. While existing TTS methods operate in a discrete token
space by generating more intermediate steps, recent studies in Coconut and
SoftCoT have demonstrated that thinking in the continuous latent space can
further enhance the reasoning performance. Such latent thoughts encode
informative thinking without the information loss associated with
autoregressive token generation, sparking increased interest in
continuous-space reasoning. Unlike discrete decoding, where repeated sampling
enables exploring diverse reasoning paths, latent representations in continuous
space are fixed for a given input, which limits diverse exploration, as all
decoded paths originate from the same latent thought. To overcome this
limitation, we introduce SoftCoT++ to extend SoftCoT to the Test-Time Scaling
paradigm by enabling diverse exploration of thinking paths. Specifically, we
perturb latent thoughts via multiple specialized initial tokens and apply
contrastive learning to promote diversity among soft thought representations.
Experiments across five reasoning benchmarks and two distinct LLM architectures
demonstrate that SoftCoT++ significantly boosts SoftCoT and also outperforms
SoftCoT with self-consistency scaling. Moreover, it shows strong compatibility
with conventional scaling techniques such as self-consistency. Source code is
available at https://github.com/xuyige/SoftCoT.


SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoning

  Large language models (LLMs) have demonstrated strong performance across a
wide range of programming tasks, yet their potential for code optimization
remains underexplored. This work investigates whether LLMs can optimize the
performance of assembly code, where fine-grained control over execution enables
improvements that are difficult to express in high-level languages. We present
a reinforcement learning framework that trains LLMs using Proximal Policy
Optimization (PPO), guided by a reward function that considers both functional
correctness, validated through test cases, and execution performance relative
to the industry-standard compiler gcc -O3. To support this study, we introduce
a benchmark of 8,072 real-world programs. Our model, Qwen2.5-Coder-7B-PPO,
achieves 96.0% test pass rates and an average speedup of 1.47x over the gcc -O3
baseline, outperforming all 20 other models evaluated, including
Claude-3.7-sonnet. These results indicate that reinforcement learning can
unlock the potential of LLMs to serve as effective optimizers for assembly code
performance.


Improving Assembly Code Performance with Large Language Models via
  Reinforcement Learning

  A key task in Artificial Intelligence is learning effective policies for
controlling agents in unknown environments to optimize performance measures.
Off-policy learning methods, like Q-learning, allow learners to make optimal
decisions based on past experiences. This paper studies off-policy learning
from biased data in complex and high-dimensional domains where \emph{unobserved
confounding} cannot be ruled out a priori. Building on the well-celebrated Deep
Q-Network (DQN), we propose a novel deep reinforcement learning algorithm
robust to confounding biases in observed data. Specifically, our algorithm
attempts to find a safe policy for the worst-case environment compatible with
the observations. We apply our method to twelve confounded Atari games, and
find that it consistently dominates the standard DQN in all games where the
observed input to the behavioral and target policies mismatch and unobserved
confounders exist.


Automatic Reward Shaping from Confounded Offline Data

  We introduce finite mixtures of Ising models as a novel approach to study
multivariate patterns of associations of binary variables. Our proposed models
combine the strengths of Ising models and multivariate Bernoulli mixture
models. We examine conditions required for the identifiability of Ising mixture
models, and develop a Bayesian framework for fitting them. Through simulation
experiments and real data examples, we show that Ising mixture models lead to
meaningful results for sparse binary contingency tables.


Bayesian Finite Mixtures of Ising Models

  Audio-Visual Video Parsing (AVVP) entails the challenging task of localizing
both uni-modal events (i.e., those occurring exclusively in either the visual
or acoustic modality of a video) and multi-modal events (i.e., those occurring
in both modalities concurrently). Moreover, the prohibitive cost of annotating
training data with the class labels of all these events, along with their start
and end times, imposes constraints on the scalability of AVVP techniques unless
they can be trained in a weakly-supervised setting, where only
modality-agnostic, video-level labels are available in the training data. To
this end, recently proposed approaches seek to generate segment-level
pseudo-labels to better guide model training. However, the absence of
inter-segment dependencies when generating these pseudo-labels and the general
bias towards predicting labels that are absent in a segment limit their
performance. This work proposes a novel approach towards overcoming these
weaknesses called Uncertainty-weighted Weakly-supervised Audio-visual Video
Parsing (UWAV). Additionally, our innovative approach factors in the
uncertainty associated with these estimated pseudo-labels and incorporates a
feature mixup based training regularization for improved training. Empirical
results show that UWAV outperforms state-of-the-art methods for the AVVP task
on multiple metrics, across two different datasets, attesting to its
effectiveness and generalizability.


UWAV: Uncertainty-weighted Weakly-supervised Audio-Visual Video Parsing

  Large intelligent surface (LIS) is one promising path to leverage 6G
performance in sub-10 GHz bands. This two-part paper explores the properties of
phase-conjugation focusing for a simplified LIS setup with a two-dimensional
(2D) circular antenna array and a user antenna located within the array
aperture in the same plane. In Part I of this article, we assume vertical
polarization for all antenna elements, whereas Part II assumes horizontal
polarization. In Part I, we focus on the effect of array radius on the peak
gain, 3 dB focusing width, and sidelobes for two types of circular arrays. The
numerical results show that the gain minimum is located at the array center.
The peak gain varies by less than 0.5 dB for focal points located within 2 from
the array center. Similarly, the focal width and sidelobe level are also stable
within this region, irrespective of array radius. From 2 from the center to the
array edge, the closer proximity of the focal points to some array elements
than other elements results in more drastic changes in these NFF properties.
Finally, full-wave simulation using Ansys HFSS is used to partially validate
the numerical results.


On Properties of Phase-Conjugation Focusing for Large Intelligent
  Surface Applications -- Part I: Vertical Polarization

  We give a complete characterization of all eternal solutions $b(x,t)$ of the
KPZ fixed point satisfying the asymptotic slope condition $\lim_{|x| \to
\infty} \frac{b(x,0)}{x} = 2\xi$. For fixed $\xi$, there is exactly one eternal
solution with probability one. However, in the second and third authors' work
with Sepp\"al\"ainen, it was shown that there exists a random, countably
infinite set of slopes, for which there exist at least two eternal solutions.
These correspond to two non-coalescing families of infinite geodesics in the
same direction for the directed landscape. We denote the two eternal solutions
as $b^{\xi-}$ and $b^{\xi +}$. In the present paper, we show that, for the
exceptional slopes, there are in fact uncountably many eternal solutions. To
give the characterization, we show that these eternal solutions are in
bijection with a certain set of bi-infinite competition interfaces. Each
bi-infinite interface separates the plane into two connected components--a left
component and a right component. A general eternal solution with slope $\xi$ is
equal to $b^{\xi-}$ on the left component and equal to $b^{\xi +}$ on the right
component. For these bi-infinite interfaces in the exceptional directions, we
uncover new geometric phenomena that is not present for directed landscape
geodesics.


Exceptional force, uncountably many solutions in the KPZ fixed point

  Cross-domain aspect-based sentiment analysis (ABSA) aims to perform various
fine-grained sentiment analysis tasks on a target domain by transferring
knowledge from a source domain. Since labeled data only exists in the source
domain, a model is expected to bridge the domain gap for tackling cross-domain
ABSA. Though domain adaptation methods have proven to be effective, most of
them are based on a discriminative model, which needs to be specifically
designed for different ABSA tasks. To offer a more general solution, we propose
a unified bidirectional generative framework to tackle various cross-domain
ABSA tasks. Specifically, our framework trains a generative model in both
text-to-label and label-to-text directions. The former transforms each task
into a unified format to learn domain-agnostic features, and the latter
generates natural sentences from noisy labels for data augmentation, with which
a more accurate model can be trained. To investigate the effectiveness and
generality of our framework, we conduct extensive experiments on four
cross-domain ABSA tasks and present new state-of-the-art results on all tasks.
Our data and code are publicly available at
\url{https://github.com/DAMO-NLP-SG/BGCA}.
