Searcharxiv⌕ Search

arXiv subjects

Victor Campos

Publications and source records attributed to Victor Campos.

At least 19 recordsLinked to original sources
arXiv2025

Understanding the medium-like effects in the jet-like yield in pp and p-Pb collisions using event generators

To understand the dynamics of jet-medium interaction in small systems such as proton-proton (pp) and proton-lead (p-Pb) collisions at $\sqrt{s_{\rm NN}}$ = 5.02 TeV, particle production is studied in three distinct topological regions defined with respect to the charged particle with the highest transverse momentum in the event ($p_{\rm T}^{\rm trig}$). The jet-like yield is defined by the particle density in the toward region ($|Δφ|<π/3$) after subtracting that in the transverse region ($π/3<|Δφ|<2π/3$). The activity on the transverse side is used as a proxy for medium-like effects. Three different Monte Carlo event generators -- \textsc{Pythia8}, a multiphase transport (AMPT) model, and EPOS4 -- are employed to investigate particle yields as a function of $p_{\rm T}^{\rm trig}$ in the interval 0.5-20\,GeV/$c$. Calculations are performed for the $p_{\rm T}$ threshold of 0.5 GeV/$c$ at mid-rapidity ($|η| < 0.8$). The jet-like yield in the toward region for pp collisions show interesting dynamics; they are significantly affected by the medium-like effects in the low to intermediate $p_{\rm T}^{\rm trig}$ ($<8$\,GeV/$c$) which is studied through color reconnection and hydrodynamics in \textsc{Pythia8} and EPOS4, respectively. However, the results from AMPT show that the jet-like yield is medium-like modified throughout the entire $p_{\rm T}^{\rm trig}$ range. The jet-like yield in p-Pb collisions using AMPT is also studied. Notably, a dip structure that is observed in the jet-like signal ratio of pp to p-Pb at low $p_{\rm T}^{\rm trig}$ in ALICE data, is reproduced by AMPT model with string melting on, pointing to possible medium-like behavior in small systems. The results of this article also underscore the importance of high-$p_{\rm T}^{\rm trig}$ ($p_{\rm T}^{\rm trig} >$ 8 GeV/$c$) for minimizing underlying event biases in jet-related studies.

hep-ph↗
arXiv2024

Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

cs.CL↗
arXiv2023

New Menger-like dualities in digraphs and applications to half-integral linkages

We present new min-max relations in digraphs between the number of paths satisfying certain conditions and the order of the corresponding cuts. We define these objects in order to capture, in the context of solving the half-integral linkage problem, the essential properties needed for reaching a large bramble of congestion two (or any other constant) from the terminal set. This strategy has been used ad-hoc in several articles, usually with lengthy technical proofs, and our objective is to abstract it to make it applicable in a simpler and unified way. We provide two proofs of the min-max relations, one consisting in applying Menger's Theorem on appropriately defined auxiliary digraphs, and an alternative simpler one using matroids, however with worse polynomial running time. As an application, we manage to simplify and improve several results of Edwards et al. [ESA 2017] and of Giannopoulou et al. [SODA 2022] about finding half-integral linkages in digraphs. Concerning the former, besides being simpler, our proof provides an almost optimal bound on the strong connectivity of a digraph for it to be half-integrally feasible under the presence of a large bramble of congestion two (or equivalently, if the directed tree-width is large, which is the hard case). Concerning the latter, our proof uses brambles as rerouting objects instead of cylindrical grids, hence yielding much better bounds and being somehow independent of a particular topology. We hope that our min-max relations will find further applications as, in our opinion, they are simple, robust, and versatile to be easily applicable to different types of routing problems in digraphs.

math.CO↗
arXiv2022

Adapting the Directed Grid Theorem into an FPT Algorithm

The Grid Theorem of Robertson and Seymour [JCTB, 1986], is one of the most important tools in the field of structural graph theory, finding numerous applications in the design of algorithms for undirected graphs. An analogous version of the Grid Theorem in digraphs was conjectured by Johnson et al. [JCTB, 2001], and proved by Kawarabayashi and Kreutzer [STOC, 2015]. Namely, they showed that there is a function $f(k)$ such that every digraph of directed tree-width at least $f(k)$ contains a cylindrical grid of size $k$ as a butterfly minor and stated that their proof can be turned into an XP algorithm, with parameter $k$, that either constructs a decomposition of the appropriate width, or finds the claimed large cylindrical grid as a butterfly minor. In this paper, we adapt some of the steps of the proof of Kawarabayashi and Kreutzer to improve this XP algorithm into an FPT algorithm. Towards this, our main technical contributions are two FPT algorithms with parameter $k$. The first one either produces an arboreal decomposition of width $3k-2$ or finds a haven of order $k$ in a digraph $D$, improving on the original result for arboreal decompositions by Johnson et al. The second algorithm finds a well-linked set of order $k$ in a digraph $D$ of large directed tree-width. As tools to prove these results, we show how to solve a generalized version of the problem of finding balanced separators for a given set of vertices $T$ in FPT time with parameter $|T|$, a result that we consider to be of its own interest.

cs.DS↗
arXiv2020

Cycle convexity and the tunnel number of links

In this work, we introduce a new graph convexity, that we call Cycle Convexity, motivated by related notions in Knot Theory. For a graph $G=(V,E)$, define the interval function in the Cycle Convexity as $I_{cc}(S) = S\cup \{v\in V(G)\mid \text{there is a cycle }C\text{ in }G\text{ such that } V(C)\setminus S=\{v\}\}$, for every $S\subseteq V(G)$. We say that $S\subseteq V(G)$ is convex if $I_{cc}(S)=S$. The convex hull of $S\subseteq V(G)$, denoted by $Hull(S)$, is the inclusion-wise minimal convex set $S'$ such that $S\subseteq S'$. A set $S\subseteq V(G)$ is called a hull set if $Hull(S)=V(G)$. The hull number of $G$ in the cycle convexity, denoted by $hn_{cc}(G)$, is the cardinality of a smallest hull set of $G$. We first present the motivation for introducing such convexity and the study of its related hull number. Then, we prove that: the hull number of a 4-regular planar graph is at most half of its vertices; computing the hull number of a planar graph is an $NP$-complete problem; computing the hull humber of chordal graphs, $P_4$-sparse graphs and grids can be done in polynomial time.

math.CO↗
arXiv2020

Edge-Disjoint Branchings in Temporal Graphs

A temporal digraph ${\cal G}$ is a triple $(G, γ, λ)$ where $G$ is a digraph, $γ$ is a function on $V(G)$ that tells us the timestamps when a vertex is active, and $λ$ is a function on $E(G)$ that tells for each $uv \in E(G)$ when $u$ and $v$ are linked. Given a static digraph $G$, and a subset $R\subseteq V(G)$, a spanning branching with root $R$ is a subdigraph of $G$ that has exactly one path from $R$ to each $v\in V(G)$. In this paper, we consider the temporal version of Edmonds' classical result about the problem of finding $k$ edge-disjoint spanning branchings respectively rooted at given $R_1,\cdots,R_k$. We introduce and investigate different definitions of spanning branchings, and of edge-disjointness in the context of temporal graphs. A branching ${\cal B}$ is vertex-spanning if the root is able to reach each vertex $v$ of $G$ at some time where $v$ is active, while it is temporal-spanning if $v$ can be reached from the root at every time where $v$ is active. On the other hand, two branchings ${\cal B}_1$ and ${\cal B}_2$ are edge-disjoint if they do not use the same edge of $G$, and are temporal-edge-disjoint if they can use the same edge of $G$ but at different times. This lead us to four definitions of disjoint spanning branchings and we prove that, unlike the static case, only one of these can be computed in polynomial time, namely the temporal-edge-disjoint temporal-spanning branchings problem, while the other versions are $\mathsf{NP}$-complete, even under very strict assumptions.

cs.DS↗
arXiv2019

How to Initialize your Network? Robust Initialization for WeightNorm & ResNets

Residual networks (ResNet) and weight normalization play an important role in various deep learning applications. However, parameter initialization strategies have not been studied previously for weight normalized networks and, in practice, initialization methods designed for un-normalized networks are used as a proxy. Similarly, initialization for ResNets have also been studied for un-normalized networks and often under simplified settings ignoring the shortcut connection. To address these issues, we propose a novel parameter initialization strategy that avoids explosion/vanishment of information across layers for weight normalized networks with and without residual connections. The proposed strategy is based on a theoretical analysis using mean field approximation. We run over 2,500 experiments and evaluate our proposal on image datasets showing that the proposed initialization outperforms existing initialization methods in terms of generalization performance, robustness to hyper-parameter values and variance between seeds, especially when networks get deeper in which case existing methods fail to even start training. Finally, we show that using our initialization in conjunction with learning rate warmup is able to reduce the gap between the performance of weight normalized and batch normalized networks.

stat.ML↗
arXiv2019

Recurrent Neural Networks for Semantic Instance Segmentation

We present a recurrent model for semantic instance segmentation that sequentially generates binary masks and their associated class probabilities for every object in an image. Our proposed system is trainable end-to-end from an input image to a sequence of labeled masks and, compared to methods relying on object proposals, does not require post-processing steps on its output. We study the suitability of our recurrent model on three different instance segmentation benchmarks, namely Pascal VOC 2012, CVPPP Plant Leaf Segmentation and Cityscapes. Further, we analyze the object sorting patterns generated by our model and observe that it learns to follow a consistent pattern, which correlates with the activations learned in the encoder part of our network. Source code and models are available at https://imatge-upc.github.io/rsis/

cs.CV↗
arXiv2018

Skip RNN: Learning to Skip State Updates in Recurrent Neural Networks

Recurrent Neural Networks (RNNs) continue to show outstanding performance in sequence modeling tasks. However, training RNNs on long sequences often face challenges like slow inference, vanishing gradients and difficulty in capturing long term dependencies. In backpropagation through time settings, these issues are tightly coupled with the large, sequential computational graph resulting from unfolding the RNN in time. We introduce the Skip RNN model which extends existing RNN models by learning to skip state updates and shortens the effective size of the computational graph. This model can also be encouraged to perform fewer state updates through a budget constraint. We evaluate the proposed model on various tasks and show how it can reduce the number of required RNN updates while preserving, and sometimes even improving, the performance of the baseline RNN models. Source code is publicly available at https://imatge-upc.github.io/skiprnn-2017-telecombcn/ .

cs.AI↗
arXiv2017

More cat than cute? Interpretable Prediction of Adjective-Noun Pairs

The increasing availability of affect-rich multimedia resources has bolstered interest in understanding sentiment and emotions in and from visual content. Adjective-noun pairs (ANP) are a popular mid-level semantic construct for capturing affect via visually detectable concepts such as "cute dog" or "beautiful landscape". Current state-of-the-art methods approach ANP prediction by considering each of these compound concepts as individual tokens, ignoring the underlying relationships in ANPs. This work aims at disentangling the contributions of the `adjectives' and `nouns' in the visual prediction of ANPs. Two specialised classifiers, one trained for detecting adjectives and another for nouns, are fused to predict 553 different ANPs. The resulting ANP prediction model is more interpretable as it allows us to study contributions of the adjective and noun components. Source code and models are available at https://imatge-upc.github.io/affective-2017-musa2/ .

cs.CV↗
arXiv2017

Disentangling Motion, Foreground and Background Features in Videos

This paper introduces an unsupervised framework to extract semantically rich features for video representation. Inspired by how the human visual system groups objects based on motion cues, we propose a deep convolutional neural network that disentangles motion, foreground and background information. The proposed architecture consists of a 3D convolutional feature encoder for blocks of 16 frames, which is trained for reconstruction tasks over the first and last frames of the sequence. A preliminary supervised experiment was conducted to verify the feasibility of proposed method by training the model with a fraction of videos from the UCF-101 dataset taking as ground truth the bounding boxes around the activity regions. Qualitative results indicate that the network can successfully segment foreground and background in videos as well as update the foreground appearance based on disentangled motion features. The benefits of these learned features are shown in a discriminative classification task, where initializing the network with the proposed pretraining method outperforms both random initialization and autoencoder pretraining. Our model and source code are publicly available at https://imatge-upc.github.io/unsupervised-2017-cvprw/ .

cs.CV↗
arXiv2017

From Pixels to Sentiment: Fine-tuning CNNs for Visual Sentiment Prediction

Visual multimedia have become an inseparable part of our digital social lives, and they often capture moments tied with deep affections. Automated visual sentiment analysis tools can provide a means of extracting the rich feelings and latent dispositions embedded in these media. In this work, we explore how Convolutional Neural Networks (CNNs), a now de facto computational machine learning tool particularly in the area of Computer Vision, can be specifically applied to the task of visual sentiment prediction. We accomplish this through fine-tuning experiments using a state-of-the-art CNN and via rigorous architecture analysis, we present several modifications that lead to accuracy improvements over prior art on a dataset of images from a popular social media platform. We additionally present visualizations of local patterns that the network learned to associate with image sentiment for insight into how visual positivity (or negativity) is perceived by the model.

cs.CV↗
arXiv2015

Edge-b-coloring Trees

A b-coloring of the vertices of a graph is a proper coloring where each color class contains a vertex which is adjacent to at least one vertex in each other color class. The b-chromatic number of $G$ is the maximum integer $b(G)$ for which $G$ has a b-coloring with $b(G)$ colors. This problem was introduced by Irving and Manlove in 1999, where they showed that computing $b(G)$ is $\mathcal{NP}$-hard in general and polynomial-time solvable for trees. Since then, a number of complexity results were shown, including NP-hardness results for chordal graphs (Havet et. al., 2011) and line graphs (Campos et. al., 2015). In this article, we present a polynomial time algorithm that solves the problem restricted to claw-free block graphs, an important subclass of chordal graphs and line graphs. This is equivalent to solving the edge coloring version of the problem restricted to trees.

cs.DM↗
arXiv2015

Polyhedral studies of vertex coloring problems: The asymmetric representatives formulation

Despite the fact that some vertex coloring problems are polynomially solvable on certain graph classes, most of these problems are not "under control" from a polyhedral point of view. The equivalence between \emph{optimization} and \emph{polyhedral separation} suggests that, for these problems, there must exist formulations admitting some elegant characterization for the polytopes associated to them. Therefore, it is interesting to study known formulations for vertex coloring with the goal of finding such characterizations. In this work we study the asymmetric representatives formulation and we show that the corresponding coloring polytope, for a given graph $G$, can be interpreted as the stable set polytope of another graph obtained from $G$. This result allows us to derive complete characterizations for the corresponding coloring polytope for some families of graphs, based on known complete characterizations for the stable set polytope.

math.CO↗
arXiv2015

Diving Deep into Sentiment: Understanding Fine-tuned CNNs for Visual Sentiment Prediction

Visual media are powerful means of expressing emotions and sentiments. The constant generation of new content in social networks highlights the need of automated visual sentiment analysis tools. While Convolutional Neural Networks (CNNs) have established a new state-of-the-art in several vision problems, their application to the task of sentiment analysis is mostly unexplored and there are few studies regarding how to design CNNs for this purpose. In this work, we study the suitability of fine-tuning a CNN for visual sentiment prediction as well as explore performance boosting techniques within this deep learning setting. Finally, we provide a deep-dive analysis into a benchmark, state-of-the-art network architecture to gain insight about how to design patterns for CNNs on the task of visual sentiment prediction.

cs.MM↗
arXiv2012

b-coloring graphs with large girth

A b-coloring of a graph is a coloring of its vertices such that every color class contains a vertex that has a neighbor in all other classes. The b-chromatic number of a graph is the largest integer k such that the graph has a b-coloring with k colors. We show how to compute in polynomial time the b-chromatic number of a graph of girth at least 9. This improves the seminal result of Irving and Manlove on trees.

math.CO↗
arXiv2011

Fixed parameter algorithms for restricted coloring problems

In this paper, we obtain polynomial time algorithms to determine the acyclic chromatic number, the star chromatic number, the Thue chromatic number, the harmonious chromatic number and the clique chromatic number of $P_4$-tidy graphs and $(q,q-4)$-graphs, for every fixed $q$. These classes include cographs, $P_4$-sparse and $P_4$-lite graphs. All these coloring problems are known to be NP-hard for general graphs. These algorithms are fixed parameter tractable on the parameter $q(G)$, which is the minimum $q$ such that $G$ is a $(q,q-4)$-graph. We also prove that every connected $(q,q-4)$-graph with at least $q$ vertices is 2-clique-colorable and that every acyclic coloring of a cograph is also nonrepetitive.

cs.DM↗