Opaque objects reconstructed by 3DGaussian Splatting (3DGS) often exhibit a falsely transparent surface, leading to inconsistent background and internal patterns under camera motion in interactive viewing. This issue stems from the ill-posed optimization in 3DGS. During training, background and foreground Gaussians are blended via α-compositing and optimized solely against the input RGB images using a photometric loss. As this process lacks an explicit constraint on surface opacity, the optimization may incorrectly assign transparency to opaque regions, resulting in view-inconsistent and falsely transparent output. This issue is difficult to detect in standard evaluation settings (i.e., rendering static images), but becomes particularly evident in object-centric reconstructions under interactive viewing.

Multimodal large language models (LLMs) have made rapid progress in visual understanding, yet their extension from images to videos often reduces to a naive concatenation of frame tokens. In this work, we investigate what video finetuning brings to multimodal LLMs. We propose Visual Chain-of-Thought (vCoT), an explicit reasoning process that generates transitional event descriptions between consecutive frames. Using vCoT, we systematically compare image-only LVLMs with their video-finetuned counterparts, both with and without access to these transitional cues. Our experiments show that vCoT significantly improves the performance of image-only models on long-form video question answering, while yielding only marginal gains for video-finetuned models. This suggests that the latter already capture frame-to-frame transitions implicitly. Moreover, we find that video models transfer this temporal reasoning ability to purely static settings, outperforming image models' baselines on relational visual reasoning tasks.

Surrogate-assisted evolutionary algorithms (SAEAs) are a key tool for addressing costly optimization tasks, with their efficiency being heavily dependent on the selection of surrogate models and infill sampling criteria. However, designing an effective dynamic selection strategy for SAEAs is labor-intensive and requires substantial domain knowledge. To address this challenge, this paper proposes LLM-SAEA, a novel approach that integrates large language models (LLMs) to configure both surrogate models and infill sampling criteria online. Specifically, LLM-SAEA develops a collaboration-of-experts framework, where one LLM serves as a scoring expert (LLM-SE), assigning scores to surrogate models and infill sampling criteria based on their optimization performance, while another LLM acts as a decision expert (LLM-DE), selecting the appropriate configurations by analyzing their scores along with the current optimization state. Experimental results demonstrate that LLM-SAEA outperforms several state-of-the-art algorithms across standard test cases. The source code is publicly available at https://github.com/ForrestXie9/LLM-SAEA.

Language models have made significant progress in generating coherent text and predicting next tokens based on input prompts. This study compares the next-token prediction performance of two well-known models: OpenAI's GPT-2 and Meta's Llama-2-7b-chat-hf on Theory of Mind (ToM) tasks. To evaluate their capabilities, we built a dataset from 10 short stories sourced from the Explore ToM Dataset. We enhanced these stories by programmatically inserting additional sentences (infills) using GPT-4, creating variations that introduce different levels of contextual complexity. This setup enables analysis of how increasing context affects model performance. We tested both models under four temperature settings (0.01, 0.5, 1.0, 2.0) and evaluated their ability to predict the next token across three reasoning levels. Zero-order reasoning involves tracking the state, either current (ground truth) or past (memory). First-order reasoning concerns understanding another's mental state (e.g., "Does Anne know the apple is salted?"). Second-order reasoning adds recursion (e.g., "Does Anne think that Charles knows the apple is salted?"). Our results show that adding more infill sentences slightly reduces prediction accuracy, as added context increases complexity and ambiguity. Llama-2 consistently outperforms GPT-2 in prediction accuracy, especially at lower temperatures, demonstrating greater confidence in selecting the most probable token. As reasoning complexity rises, model responses diverge more. Notably, GPT-2 and Llama-2 display greater variability in predictions during first- and second-order reasoning tasks. These findings illustrate how model architecture, temperature, and contextual complexity influence next-token prediction, contributing to a better understanding of the strengths and limitations of current language models.

The exploration of novel architectures requires physics-based simulation due to a lack of prior experience to start from, which introduces two specific challenges for optimization algorithms: evaluations become more expensive (in time) and evaluations might fail. The former challenge is addressed by Surrogate-Based Optimization (SBO) algorithms, in particular Bayesian Optimization (BO) using Gaussian Process (GP) models. An overview is provided of how BO can deal with challenges specific to architecture optimization, such as design variable hierarchy and multiple objectives: specific measures include ensemble infills and a hierarchical sampling algorithm. Evaluations might fail due to non-convergence of underlying solvers or infeasible geometry in certain areas of the design space. Such failed evaluations, also known as hidden constraints, pose a particular challenge to SBO/BO, as the surrogate model cannot be trained on empty results. This work investigates various strategies for satisfying hidden constraints in BO algorithms. Three high-level strategies are identified: rejection of failed points from the training set, replacing failed points based on viable (non-failed) points, and predicting the failure region. Through investigations on a set of test problems including a jet engine architecture optimization problem, it is shown that best performance is achieved with a mixed-discrete GP to predict the Probability of Viability (PoV), and by ensuring selected infill points satisfy some minimum PoV threshold. This strategy is demonstrated by solving a jet engine architecture problem that features at 50% failure rate and could not previously be solved by a BO algorithm. The developed BO algorithm and used test problems are available in the open-source Python library SBArchOpt.

This article introduces an artistic research project that utilises artist-in-residency and exhibition as methods for exploring the possibilities of robotic 3D printing and ceramics. The interdisciplinary project unites artists and architects to collaborate on a proposed curatorial concept and Do-It-With-Others (DIWO) technological development. Constraints include material, specifically local clay, production technique, namely 3D printing with a robotic arm, and kiln size, as well as an exhibition concept that is further elaborated in the next chapter. The pictorial presents four projects as case studies demonstrating how the creatives integrate these constraints into their processes. This integration leads to the subsequent refinement and customization of the robotic-ceramics interface, aligning with the practitioners' requirements through software development. The project's focus extends beyond artistic outcomes, aiming also to advance the pipeline of 3D robotic printing in clay, employing a digitally controlled material press that has been developed in-house, with its functionality refined through practice.

While existing code large language models (code LLMs) exhibit impressive capabilities in code generation, their autoregressive sequential generation inherently lacks reversibility. This limitation hinders them from timely correcting previous missing statements during coding as humans do, often leading to error propagation and suboptimal performance. We introduce JumpCoder, a novel modelagnostic framework that enables online modification and non-sequential generation to augment the code LLMs. The key idea behind JumpCoder is to insert new code into the currently generated code when necessary during generation, which is achieved through an auxiliary infilling model that works in tandem with the code LLM. Since identifying the best infill position beforehand is intractable, we adopt an infill-first, judge-later strategy, which experiments with filling at the $k$ most critical positions following the generation of each line, and uses an Abstract Syntax Tree (AST) parser alongside the Generation Model Scoring to effectively judge the validity of each potential infill. Extensive experiments using six state-of-the-art code LLMs across multiple benchmarks consistently indicate significant improvements over all baselines. Notably, JumpCoder assists code LLMs in achieving up to a 3.6% increase in Pass@1 for Python, 6.3% for Java, and 3.7% for C++ in the multilingual HumanEval benchmarks. Our code is public at https://github.com/Keytoyze/JumpCoder.

Land use / land cover (LULC) modeling is a challenging task due to long-range dependencies between geographic features and distinct spatial patterns related to topography, ecology, and human development. We identify a close connection between modeling of spatial patterns of land use and the task of image inpainting from computer vision and conduct a study of a modified PixelCNN architecture with approximately 19 million parameters for modeling LULC. In comparison with a benchmark spatial statistical model, we find that the former is capable of capturing much richer spatial correlation patterns such as roads and water bodies but does not produce a calibrated predictive distribution, suggesting the need for additional tuning. We find evidence of predictive underdispersion with regard to important ecologically-relevant land use statistics such as patch count and adjacency which can be ameliorated to some extent by manipulating sampling variability.

Large language models (LLMs) have demonstrated remarkable abilities in representation learning for program synthesis and understanding tasks. The quality of the learned representations appears to be dictated by the neural scaling laws as a function of the number of model parameters and observations, while imposing upper bounds on the model performance by the amount of available data and compute, which is costly. In this study, we attempt to render the training of LLMs for program synthesis more efficient by unifying four key components: (1) model architectures, (2) learning methods, (3) infill sampling, and, (4) data distributions. Specifically, for the model architecture, we attempt to unify encoder and decoder-based models into a single prefix-LM. For learning methods, (i) causal language modeling, (ii) span corruption, (iii) infilling are unified into a simple learning algorithm. For infill sampling, we explore the claim of a "free lunch" hypothesis. For data distributions, the effect of a mixture distribution and multi-epoch training of programming and natural languages on model performance is explored. We conduct a comprehensive series of empirical experiments on 1B LLMs, for which failures and successes of this exploration are distilled into five lessons. We will provide a final recipe for training and release CodeGen2 models in size 1B, 3.7B, 7B, and, 16B parameters, along with the training framework as open-source: https://github.com/salesforce/CodeGen.

Code is seldom written in a single left-to-right pass and is instead repeatedly edited and refined. We introduce InCoder, a unified generative model that can perform program synthesis (via left-to-right generation) as well as editing (via infilling). InCoder is trained to generate code files from a large corpus of permissively licensed code, where regions of code have been randomly masked and moved to the end of each file, allowing code infilling with bidirectional context. Our model is the first generative model that is able to directly perform zero-shot code infilling, which we evaluate on challenging tasks such as type inference, comment generation, and variable re-naming. We find that the ability to condition on bidirectional context substantially improves performance on these tasks, while still performing comparably on standard program synthesis benchmarks in comparison to left-to-right only models pretrained at similar scale. The InCoder models and code are publicly released. https://sites.google.com/view/incoder-code-models