Search

We introduce the Multi-Motion Discrete Diffusion Models (M2D2M), a novel approach for human motion generation from textual descriptions of multiple actions, utilizing the strengths of discrete diffusion models. This approach adeptly addresses the challenge of generating multi-motion sequences, ensuring seamless transitions of motions and coherence across a series of actions. The strength of M2D2M lies in its dynamic transition probability within the discrete diffusion model, which adapts transition probabilities based on the proximity between motion tokens, encouraging mixing between different modes. Complemented by a two-phase sampling strategy that includes independent and joint denoising steps, M2D2M effectively generates long-term, smooth, and contextually coherent human motion sequences, utilizing a model trained for single-motion generation. Extensive experiments demonstrate that M2D2M surpasses current state-of-the-art benchmarks for motion generation from text descriptions, showcasing its efficacy in interpreting language semantics and generating dynamic, realistic motions.

We propose a novel prompt tuning method called CoAPT(Context Attribute words in Prompt Tuning) for few/zero-shot image classification. The core motivation is that attributes are descriptive words with rich information about a given concept. Thus, we aim to enrich text queries of existing prompt tuning methods, improving alignment between text and image embeddings in CLIP embedding space. To do so, CoAPT integrates attribute words as additional prompts within learnable prompt tuning and can be easily incorporated into various existing prompt tuning methods. To facilitate the incorporation of attributes into text embeddings and the alignment with image embeddings, soft prompts are trained together with an additional meta-network that generates input-image-wise feature biases from the concatenated feature encodings of the image-text combined queries. Our experiments demonstrate that CoAPT leads to considerable improvements for existing baseline methods on several few/zero-shot image classification tasks, including base-to-novel generalization, cross-dataset transfer, and domain generalization. Our findings highlight the importance of combining hard and soft prompts and pave the way for future research on the interplay between text and image latent spaces in pre-trained models., Comment: 14 pages, 4 figures

Predicting future trajectories for other road agents is an essential task for autonomous vehicles. Established trajectory prediction methods primarily use agent tracks generated by a detection and tracking system and HD map as inputs. In this work, we propose a novel method that also incorporates visual input from surround-view cameras, allowing the model to utilize visual cues such as human gazes and gestures, road conditions, vehicle turn signals, etc, which are typically hidden from the model in prior methods. Furthermore, we use textual descriptions generated by a Vision-Language Model (VLM) and refined by a Large Language Model (LLM) as supervision during training to guide the model on what to learn from the input data. Despite using these extra inputs, our method achieves a latency of 53 ms, making it feasible for real-time processing, which is significantly faster than that of previous single-agent prediction methods with similar performance. Our experiments show that both the visual inputs and the textual descriptions contribute to improvements in trajectory prediction performance, and our qualitative analysis highlights how the model is able to exploit these additional inputs. Lastly, in this work we create and release the nuScenes-Text dataset, which augments the established nuScenes dataset with rich textual annotations for every scene, demonstrating the positive impact of utilizing VLM on trajectory prediction. Our project page is at https://moonseokha.github.io/VisionTrap/, Comment: Accepted at ECCV 2024

The analysis of high-intensity runs (or sprints) in soccer has long been a topic of interest for sports science researchers and practitioners. In particular, recent studies suggested contextualizing sprints based on their tactical purposes to better understand the physical-tactical requirements of modern match-play. However, they have a limitation in scalability, as human experts have to manually classify hundreds of sprints for every match. To address this challenge, this paper proposes a deep learning framework for automatically classifying sprints in soccer into contextual categories. The proposed model covers the permutation-invariant and sequential nature of multi-agent trajectories in soccer by deploying Set Transformers and a bidirectional GRU. We train the model with category labels made through the collaboration of human annotators and a rule-based classifier. Experimental results show that our model classifies sprints in the test dataset into 15 categories with the accuracy of 77.65%, implying the potential of the proposed framework for facilitating the integrated analysis of soccer sprints at scale., Comment: Accepted at IJCAI 2024 Workshop on Intelligent Technologies for Precision Sports Science (IT4PSS 2024)

Relative belief inferences are shown to arise as Bayes rules or limiting Bayes rules. These inferences are invariant under reparameterizations and possess a number of optimal properties. In particular, relative belief inferences are based on a direct measure of statistical evidence., Comment: arXiv admin note: substantial text overlap with arXiv:1104.3258

Effective meetings are effortful, but traditional videoconferencing systems offer little support for reducing this effort across the meeting lifecycle. Generative AI (GenAI) has the potential to radically redefine meetings by augmenting intentional meeting behaviors. CoExplorer, our novel adaptive meeting prototype, preemptively generates likely phases that meetings would undergo, tools that allow capturing attendees' thoughts before the meeting, and for each phase, window layouts, and appropriate applications and files. Using CoExplorer as a technology probe in a guided walkthrough, we studied its potential in a sample of participants from a global technology company. Our findings suggest that GenAI has the potential to help meetings stay on track and reduce workload, although concerns were raised about users' agency, trust, and possible disruption to traditional meeting norms. We discuss these concerns and their design implications for the development of GenAI meeting technology., Comment: DIS 2024

In this article, we investigate large prime factors of Fourier coefficients of non-CM normalized cuspidal Hecke eigenforms in short intervals. One of the new ingredients involves deriving an explicit version of Chebotarev density theorem in an interval of length $\frac{x}{(\log x)^A}$ for any $A>0$, modifying an earlier work of Balog and Ono. Furthermore, we need to strengthen a work of Rouse-Thorner to derive a lower bound for the largest prime factor of Fourier coefficients in an interval of length $x^{1/2 + \epsilon}$ for any $\epsilon >0$., Comment: To appear in Monatshefte f\"ur Mathematik

In a recent work arXiv:2004.14450, it has been shown that $L$-functions associated with arbitrary non-zero cusp forms take large values at the central critical point. The goal of this note is to derive analogous results for twists of Dirichlet-type functions. More precisely, for an odd integer $q >1$, let $F$ be a non-zero $\mathbb{C}$-linear combination of primitive, complex, even Dirichlet characters of conductor $q$. We show that for any $\epsilon>0$ and sufficiently large $X$, there are $\gg X^{1-\epsilon}$ fundamental discriminants $8d$ with $X < d \leq 2X$ and ${(d, 2q)=1}$ such that ${|L(1/2, F \otimes \chi_{8d})| }$ is large.

In 2008, Soundararajan showed that there exists a normalized Hecke eigenform $f$ of weight $k$ and level one such that $$ L(1/2, f ) ~\geq~ \exp\Bigg( (1 + o(1)) \sqrt{\frac{2\log k}{\log\log k} }\Bigg) $$ for sufficiently large $k \equiv 0 \pmod{4}$. In this note, we show that for any $\epsilon>0$ and for all sufficiently large $k \equiv 0 \pmod{4}$, the number of normalized Hecke eigenforms of weight $k$ and level one for which $$ L(1/2, f ) ~\geq~ \exp\left(1.41\sqrt{ \frac{ \log k }{\log\log k} }\right) $$ is $\gg_{\epsilon} k^{1-\epsilon}$. For an odd fundamental discriminant $D$, let $B_{k}(|D|)$ be the set of all cuspidal normalized Hecke eigenforms of weight $k$ and level dividing $|D|$. When the real primitive Dirichlet character $\chi_D$ satisfies $\chi_D(-1)= i^k$, we investigate the number of $f \in B_{k}(|D|)$ for which $L(1/2, f \otimes \chi_D)$ takes extremal values.

As 3D Gaussian Splatting (3DGS) provides fast and high-quality novel view synthesis, it is a natural extension to deform a canonical 3DGS to multiple frames for representing a dynamic scene. However, previous works fail to accurately reconstruct complex dynamic scenes. We attribute the failure to the design of the deformation field, which is built as a coordinate-based function. This approach is problematic because 3DGS is a mixture of multiple fields centered at the Gaussians, not just a single coordinate-based framework. To resolve this problem, we define the deformation as a function of per-Gaussian embeddings and temporal embeddings. Moreover, we decompose deformations as coarse and fine deformations to model slow and fast movements, respectively. Also, we introduce a local smoothness regularization for per-Gaussian embedding to improve the details in dynamic regions. Project page: https://jeongminb.github.io/e-d3dgs/, Comment: ECCV 2024. Project page: https://jeongminb.github.io/e-d3dgs/

Recently, a polynomial time classical algorithm has been found for learning the shallow representation of a unitary $U$ acting on $n$-qubits, by learning local inversions and then sewing them back together with ancilla qubits and SWAP gates. In this work, we bootstrap local inversion learning (LIL) to optimize quantum circuit depth by learning shallow representations for its sub-unitaries. We recursively cut circuits and apply the LIL algorithm to replace sub-circuits with their shallow representations, if it can be found by the algorithm. If not, we keep cutting until the optimization terminates, either by finding shallow representations or by reaching constant-depth sub-circuits. By replacing sub-circuits with their shallow representations, we hope to obtain some compression of the quantum circuit. Due to the binary search structure, the optimization algorithm has time complexity logarithmic in the depth of the original given circuit., Comment: 5 pages, 2 figures

Let $\tau$ denote the Ramanujan tau function. One is interested in possible prime values of $\tau$ function. Since $\tau$ is multiplicative and $\tau(n)$ is odd if and only if $n$ is an odd square, we only need to consider $\tau(p^{2n})$ for primes $p$ and natural numbers $n \geq 1$. This is a rather delicate question. In this direction, we show that for any $\epsilon > 0$ and integer $n \geq 1$, the largest prime factor of $\tau(p^{2n})$, denoted by $P(\tau(p^{2n}))$, satisfies $$ P(\tau(p^{2n})) ~>~ (\log p)^{1/8}(\log\log p)^{3/8 -\epsilon} $$ for almost all primes $p$. This improves a recent work of Bennett, Gherga, Patel and Siksek. Our results are also valid for any non-CM normalized Hecke eigenforms with integer Fourier coefficients.

In this article, we investigate a non-Archimedean analogue of a question of Atkin and Serre. More precisely, we derive lower bounds for the largest prime factor of non-zero Fourier coefficients of non-CM normalized cuspidal Hecke eigenforms of even weight $k \geq 2$, level $N$ with integer Fourier coefficients. In particular, we show that for such a form $f$ and for any real number $\epsilon>0$, the largest prime factor of the $p$-th Fourier coefficient $a_f(p)$ of $f$, denoted by $P(a_f(p))$, satisfies $$ P(a_f(p)) ~>~ (\log p)^{1/8}(\log\log p)^{3/8 -\epsilon} $$ for almost all primes $p$. This improves on earlier bounds. We also investigate a number field analogue of a recent result of Bennett, Gherga, Patel and Siksek about the largest prime factor of $a_f(p^m)$ for $m \geq 2$.

A well known result of Breulmann states that Hecke eigenvalues of Saito-Kurokawa lifts are positive. In this article, we show that the Hecke eigenvalues of an Ikeda lift at primes are positive. Further, we derive lower and upper bounds of these Hecke eigenvalues for all primes $p$. One of the main ingredients involves expressing the Hecke eigenvalues of an Ikeda lift in terms of certain reciprocal polynomials.

In Fringe Projection Profilometry (FPP), achieving robust and accurate 3D reconstruction with a limited number of fringe patterns remains a challenge in structured light 3D imaging. Conventional methods require a set of fringe images, but using only one or two patterns complicates phase recovery and unwrapping. In this study, we introduce SFNet, a symmetric fusion network that transforms two fringe images into an absolute phase. To enhance output reliability, Our framework predicts refined phases by incorporating information from fringe images of a different frequency than those used as input. This allows us to achieve high accuracy with just two images. Comparative experiments and ablation studies validate the effectiveness of our proposed method. The dataset and code are publicly accessible on our project page https://wonhoe-kim.github.io/SFNet., Comment: 35 pages, 13 figures, 6 tables

We consider a modified graphene model under exchange couplings. Various quantum anomalous phases are known to emerge under uniform or staggered exchange couplings. We introduce the twist between the orientations of two sublattice exchange couplings, which is useful for examining how such topologically nontrivial phases under different types of exchange couplings are connected to one another. The phase diagrams constructed by the variation of exchange coupling strengths and twist angles exhibit rich structures of successive topological transitions. We analyze the emergence of peculiar phases in terms of the evolution of the energy dispersions. Perturbation schemes applied to the energy levels turn out to reproduce well phase boundary lines up to moderate values of the twist angle. We also discover two close topological transitions under uniform exchange couplings, which is attributed to the interplay of the trigonal-warping deformation due to Rashba spin-orbit coupling and the staggered sublattice potential. Finally the implications of Berry curvature structure and topological excitations in real and pseudo spin textures are discussed.

Integrated computational materials engineering (ICME) has significantly enhanced the systemic analysis of the relationship between microstructure and material properties, paving the way for the development of high-performance materials. However, analyzing microstructure-sensitive material behavior remains challenging due to the scarcity of three-dimensional (3D) microstructure datasets. Moreover, this challenge is amplified if the microstructure is anisotropic, as this results in anisotropic material properties as well. In this paper, we present a framework for reconstruction of anisotropic microstructures solely based on two-dimensional (2D) micrographs using conditional diffusion-based generative models (DGMs). The proposed framework involves spatial connection of multiple 2D conditional DGMs, each trained to generate 2D microstructure samples for three different orthogonal planes. The connected multiple reverse diffusion processes then enable effective modeling of a Markov chain for transforming noise into a 3D microstructure sample. Furthermore, a modified harmonized sampling is employed to enhance the sample quality while preserving the spatial connection between the slices of anisotropic microstructure samples in 3D space. To validate the proposed framework, the 2D-to-3D reconstructed anisotropic microstructure samples are evaluated in terms of both the spatial correlation function and the physical material behavior. The results demonstrate that the framework is capable of reproducing not only the statistical distribution of material phases but also the material properties in 3D space. This highlights the potential application of the proposed 2D-to-3D reconstruction framework in establishing microstructure-property linkages, which could aid high-throughput material design for future studies

We investigate emergent topological gapless phases in the square-lattice Kitaev model with additional hopping terms. In the presence of nearest-neighbor hopping only, the model is known to exhibit gapless phases with two topological gapless points. When the strength of the newly added next-nearest-neighbor hopping is smaller than a certain value, qualitatively the same phase diagram persists. We find that further increase of the extra hopping results in a new topological phase with four gapless points. We construct a phase diagram to clarify the regions of emergent topological gapless phases as well as topologically trivial ones in the space of the chemical potential and the next-nearest-neighbor hopping strength. We examine the evolution of the gapless phases in the energy dispersions of the bulk as the chemical potential varies. The topological properties of the gapless phases are characterized by the winding numbers of the present gapless points. We also consider the ribbon geometry to examine the corresponding topological edge states. It is revealed that Majorana-fermion edge modes exist as flat bands in topological gapless phases. We also perform the analytical calculation as to Majorana-fermion zero-energy modes and discuss its implications on the numerical results.

Screening is critical for prevention and early detection of cervical cancer but it is time-consuming and laborious. Supervised deep convolutional neural networks have been developed to automate pap smear screening and the results are promising. However, the interest in using only normal samples to train deep neural networks has increased owing to class imbalance problems and high-labeling costs that are both prevalent in healthcare. In this study, we introduce a method to learn explainable deep cervical cell representations for pap smear cytology images based on one class classification using variational autoencoders. Findings demonstrate that a score can be calculated for cell abnormality without training models with abnormal samples and localize abnormality to interpret our results with a novel metric based on absolute difference in cross entropy in agglomerative clustering. The best model that discriminates squamous cell carcinoma (SCC) from normals gives 0.908 +- 0.003 area under operating characteristic curve (AUC) and one that discriminates high-grade epithelial lesion (HSIL) 0.920 +- 0.002 AUC. Compared to other clustering methods, our method enhances the V-measure and yields higher homogeneity scores, which more effectively isolate different abnormality regions, aiding in the interpretation of our results. Evaluation using in-house and additional open dataset show that our model can discriminate abnormality without the need of additional training of deep models., Comment: 20 pages, 6 figures

We study correlations between geometric subfactors living on the Ryu-Takayanagi surface that bounds the entanglement wedge. Using the surface-state correspondence and the bit threads program, we are able to calculate mutual information and conditional mutual information between subfactors. This enables us to count the shared Bell pairs between subfactors, and we propose an entanglement distillation procedure over these subsystems via a SWAP gate protocol. We comment on extending to multipartite entanglement., Comment: 10 pages, 3 figures. Version to be published, updated references