The emergence of large-scale multi-modal generative models has drastically advanced artificial intelligence, introducing unprecedented levels of performance and functionality. However, optimizing these models remains challenging due to historically isolated paths of model-centric and data-centric developments, leading to suboptimal outcomes and inefficient resource utilization. In response, we present a novel sandbox suite tailored for integrated data-model co-development. This sandbox provides a comprehensive experimental platform, enabling rapid iteration and insight-driven refinement of both data and models. Our proposed "Probe-Analyze-Refine" workflow, validated through applications on state-of-theart LLaVA-like and DiT-based models, yields significant performance boosts, such as topping the VBench leaderboard. We also uncover fruitful insights gleaned from exhaustive benchmarks, shedding light on the critical interplay between data quality, diversity, and model behavior. With the hope of fostering deeper understanding and future progress in multi-modal data and generative modeling, our codes, datasets, and models are maintained and accessible at https://github. The advent of multi-modal generative models has revolutionized artificial intelligence, pushing the boundaries of functionality and creativity across various domains (OpenAI, 2024a;b; Wang et al., 2024). Recognizing the pivotal role of training data in shaping model performance, there are fast-growing efforts to curate datasets of larger scales and higher quality (Jakubik et al., 2024). However, the development trajectories of these models and datasets have historically diverged, guided more by intuition than by systematic co-development methodologies. Recent advances in enhancing multi-modal generative models tend to be either model-centric or data-centric, rarely bridging the two aspects cohesively. For example, model-centric methods focus on algorithmic enhancements and architectural innovations under fixed data priors, while data-centric strategies usually concentrate on processing and cleaning datasets independently of specific model training contexts (Qin et al., 2024). Both approaches usually suffer from a lack of systematic guidance and cooperative synergy, relying heavily on heuristic exploration and single-perspective expertise. This fragmented landscape presents a significant barrier to achieving optimal model performance, as the interplay between data characteristics and model capabilities remains largely underexploited. Moreover, the practical implementation of multi-modal generative models is further complicated by infrastructure constraints, escalating computational costs, and the accelerating pace of development cycles (Xu et al., 2024b).

On manifolds with metrics, minimizing geodesics, or shortest paths, are minimum-length curves connecting points. Various real-world tasks can be reduced to the generation of geodesics on manifolds. Examples include time-optimal path planning on sloping ground [Matsumoto, 1989], robot motion planning under various constraints [LaValle, 2006, Ratliff et al., 2015], physical systems [Pfeifer, 2019], the Wasserstein distance [Agueh, 2012], and image morphing [Michelis and Becker, 2021, Effland et al., 2021]. Typically, metrics are only known infinitesimally (a form of a Riemannian or Finsler metric), and their distance functions are not known beforehand. Computation of geodesics by solving optimization problems or differential equations is generally computationally costly and requires an explicit form of the metric, or at least, values of its differentials.

Instruction generation is a vital and multidisciplinary research area with broad applications. Existing instruction generation models are limited to generating instructions in a single style from a particular dataset, and the style and content of generated instructions cannot be controlled. Moreover, most existing instruction generation methods also disregard the spatial modeling of the navigation environment. Leveraging the capabilities of Large Language Models (LLMs), we propose C-Instructor, which utilizes the chain-of-thought-style prompt for style-controllable and content-controllable instruction generation. Firstly, we propose a Chain of Thought with Landmarks (CoTL) mechanism, which guides the LLM to identify key landmarks and then generate complete instructions. CoTL renders generated instructions more accessible to follow and offers greater controllability over the manipulation of landmark objects. Furthermore, we present a Spatial Topology Modeling Task to facilitate the understanding of the spatial structure of the environment. Finally, we introduce a Style-Mixed Training policy, harnessing the prior knowledge of LLMs to enable style control for instruction generation based on different prompts within a single model instance. Extensive experiments demonstrate that instructions generated by C-Instructor outperform those generated by previous methods in text metrics, navigation guidance evaluation, and user studies.

Knowledge distillation (KD) involves transferring the knowledge from one neural network to another, often from a larger, well-trained model (teacher) to a smaller, more efficient model (student). Traditional KD methods minimize the Kullback-Leibler (KL) divergence between the probabilistic outputs of the teacher and student networks. However, this approach often overlooks crucial structural knowledge embedded within the teacher's network. In this paper, we introduce Invariant Consistency Distillation (ICD), a novel methodology designed to enhance KD by ensuring that the student model's representations are consistent with those of the teacher. Our approach combines contrastive learning with an explicit invariance penalty, capturing significantly more information from the teacher's representation of the data. Our results on CIFAR-100 demonstrate that ICD outperforms traditional KD techniques and surpasses 13 state-of-the-art methods. In some cases, the student model even exceeds the teacher model in terms of accuracy. Furthermore, we successfully transfer our method to other datasets, including Tiny ImageNet and STL-10. The code will be made public soon.

Sign language translation (SLT) addresses the problem of translating information from a sign language in video to a spoken language in text. Existing studies, while showing progress, are often limited to narrow domains and/or few sign languages and struggle with open-domain tasks. In this paper, we push forward the frontier of SLT by scaling pretraining data, model size, and number of translation directions. We perform large-scale SLT pretraining on different data including 1) noisy multilingual YouTube SLT data, 2) parallel text corpora, and 3) SLT data augmented by translating video captions to other languages with off-the-shelf machine translation models. We unify different pretraining tasks with task-specific prompts under the encoder-decoder architecture, and initialize the SLT model with pretrained (m/By)T5 models across model sizes. SLT pretraining results on How2Sign and FLEURS-ASL#0 (ASL to 42 spoken languages) demonstrate the significance of data/model scaling and cross-lingual cross-modal transfer, as well as the feasibility of zero-shot SLT. We finetune the pretrained SLT models on 5 downstream open-domain SLT benchmarks covering 5 sign languages. Experiments show substantial quality improvements over the vanilla baselines, surpassing the previous state-of-the-art (SOTA) by wide margins.

Knowledge distillation (KD) is an effective method for transferring knowledge from a large, well-trained teacher model to a smaller, more efficient student model. Despite its success, one of the main challenges in KD is ensuring the efficient transfer of complex knowledge while maintaining the student's computational efficiency. Unlike previous works that applied contrastive objectives promoting explicit negative instances, we introduce Relational Representation Distillation (RRD). Our approach leverages pairwise similarities to explore and reinforce the relationships between the teacher and student models. Inspired by self-supervised learning principles, it uses a relaxed contrastive loss that focuses on similarity rather than exact replication. This method aligns the output distributions of teacher samples in a large memory buffer, improving the robustness and performance of the student model without the need for strict negative instance differentiation. Our approach demonstrates superior performance on CIFAR-100, outperforming traditional KD techniques and surpassing 13 state-of-the-art methods. It also transfers successfully to other datasets like Tiny ImageNet and STL-10. The code will be made public soon.

Modelling growth in student achievement is a significant challenge in the field of education. Understanding how interventions or experiences such as part-time work can influence this growth is also important. Traditional methods like difference-in-differences are effective for estimating causal effects from longitudinal data. Meanwhile, Bayesian non-parametric methods have recently become popular for estimating causal effects from single time point observational studies. However, there remains a scarcity of methods capable of combining the strengths of these two approaches to flexibly estimate heterogeneous causal effects from longitudinal data. Motivated by two waves of data from the High School Longitudinal Study, the NCES' most recent longitudinal study which tracks a representative sample of over 20,000 students in the US, our study introduces a longitudinal extension of Bayesian Causal Forests. This model allows for the flexible identification of both individual growth in mathematical ability and the effects of participation in part-time work. Simulation studies demonstrate the predictive performance and reliable uncertainty quantification of the proposed model. Results reveal the negative impact of part time work for most students, but hint at potential benefits for those students with an initially low sense of school belonging. Clear signs of a widening achievement gap between students with high and low academic achievement are also identified. Potential policy implications are discussed, along with promising areas for future research.

We formulate well-posed continuous-time generative flows for learning distributions that are supported on low-dimensional manifolds through Wasserstein proximal regularizations of $f$-divergences. Wasserstein-1 proximal operators regularize $f$-divergences so that singular distributions can be compared. Meanwhile, Wasserstein-2 proximal operators regularize the paths of the generative flows by adding an optimal transport cost, i.e., a kinetic energy penalization. Via mean-field game theory, we show that the combination of the two proximals is critical for formulating well-posed generative flows. Generative flows can be analyzed through optimality conditions of a mean-field game (MFG), a system of a backward Hamilton-Jacobi (HJ) and a forward continuity partial differential equations (PDEs) whose solution characterizes the optimal generative flow. For learning distributions that are supported on low-dimensional manifolds, the MFG theory shows that the Wasserstein-1 proximal, which addresses the HJ terminal condition, and the Wasserstein-2 proximal, which addresses the HJ dynamics, are both necessary for the corresponding backward-forward PDE system to be well-defined and have a unique solution with provably linear flow trajectories. This implies that the corresponding generative flow is also unique and can therefore be learned in a robust manner even for learning high-dimensional distributions supported on low-dimensional manifolds. The generative flows are learned through adversarial training of continuous-time flows, which bypasses the need for reverse simulation. We demonstrate the efficacy of our approach for generating high-dimensional images without the need to resort to autoencoders or specialized architectures.

We present a unified theory for Mahalanobis-type anomaly detection on Banach spaces, using ideas from Cameron-Martin theory applied to non-Gaussian measures. This approach leads to a basis-free, data-driven notion of anomaly distance through the so-called variance norm of a probability measure, which can be consistently estimated using empirical measures. Our framework generalizes the classical $\mathbb{R}^d$, functional $(L^2[0,1])^d$, and kernelized settings, including the general case of non-injective covariance operator. We prove that the variance norm depends solely on the inner product in a given Hilbert space, and hence that the kernelized Mahalanobis distance can naturally be recovered by working on reproducing kernel Hilbert spaces. Using the variance norm, we introduce the notion of a kernelized nearest-neighbour Mahalanobis distance for semi-supervised anomaly detection. In an empirical study on 12 real-world datasets, we demonstrate that the kernelized nearest-neighbour Mahalanobis distance outperforms the traditional kernelized Mahalanobis distance for multivariate time series anomaly detection, using state-of-the-art time series kernels such as the signature, global alignment, and Volterra reservoir kernels. Moreover, we provide an initial theoretical justification of nearest-neighbour Mahalanobis distances by developing concentration inequalities in the finite-dimensional Gaussian case.

We present Cephalo, a series of multimodal vision large language models (V-LLMs) designed for materials science applications, integrating visual and linguistic data for enhanced understanding. A key innovation of Cephalo is its advanced dataset generation method. Cephalo is trained on integrated image and text data from thousands of scientific papers and science-focused Wikipedia data demonstrates can interpret complex visual scenes, generate precise language descriptions, and answer queries about images effectively. The combination of a vision encoder with an autoregressive transformer supports multimodal natural language understanding, which can be coupled with other generative methods to create an image-to-text-to-3D pipeline. To develop more capable models from smaller ones, we report both mixture-of-expert methods and model merging. We examine the models in diverse use cases that incorporate biological materials, fracture and engineering analysis, protein biophysics, and bio-inspired design based on insect behavior. Generative applications include bio-inspired designs, including pollen-inspired architected materials, as well as the synthesis of bio-inspired material microstructures from a photograph of a solar eclipse. Additional model fine-tuning with a series of molecular dynamics results demonstrate Cephalo's enhanced capabilities to accurately predict statistical features of stress and atomic energy distributions, as well as crack dynamics and damage in materials.