Zero-shot optimization involves optimizing a target task that was not seen during training, aiming to provide the optimal solution without or with minimal adjustments to the optimizer. It is crucial to ensure reliable and robust performance in various applications. Current optimizers often struggle with zero-shot optimization and require intricate hyperparameter tuning to adapt to new tasks. To address this, we propose a Pretrained Optimization Model (POM) that leverages knowledge gained from optimizing diverse tasks, offering efficient solutions to zero-shot optimization through direct application or fine-tuning with few-shot samples. Evaluation on the BBOB benchmark and two robot control tasks demonstrates that POM outperforms state-of-the-art black-box optimization methods, especially for high-dimensional tasks.

Neural Information Processing Systems http://nips.cc/

It is crucial to ensure reliable and robust performance in various applications. Current optimizers often struggle with zero-shot optimization and require intricate hyperparameter tuning to adapt to new tasks.

To this end, we pose two problems.

Zero-shot optimization involves optimizing a target task that was not seen during training, aiming to provide the optimal solution without or with minimal adjustments to the optimizer. It is crucial to ensure reliable and robust performance in various applications. Current optimizers often struggle with zero-shot optimization and require intricate hyperparameter tuning to adapt to new tasks. To address this, we propose a Pretrained Optimization Model (POM) that leverages knowledge gained from optimizing diverse tasks, offering efficient solutions to zero-shot optimization through direct application or fine-tuning with few-shot samples. Evaluation on the BBOB benchmark and two robot control tasks demonstrates that POM outperforms state-of-the-art black-box optimization methods, especially for high-dimensional tasks.

As an artistic aid in tiled level design, Constraint Based Tiling Generation (CBTG) algorithms can help to automatically create level realizations from a set of tiles and placement constraints. Merrell's Modify in Blocks Model Synthesis (MMS) and Gumin's Wave Function Collapse (WFC) have been proposed as Constraint Based Tiling Generation (CBTG) algorithms that work well for many scenarios but have limitations in problem size, problem setup and solution biasing. We present Punch Out Model Synthesis (POMS), a Constraint Based Tiling Generation algorithm, that can handle large problem sizes, requires minimal assumptions for setup and can help mitigate solution biasing. POMS attempts to resolve indeterminate grid regions by trying to progressively realize sub-blocks, performing a stochastic boundary erosion on previously resolved regions should sub-block resolution fail. We highlight the results of running a reference implementation on different tile sets and discuss a tile correlation length, implied by the tile constraints, and its role in choosing an appropriate block size to aid POMS in successfully finding grid realizations.

Diffusion models based on Multi-Head Attention (MHA) have become ubiquitous to generate high quality images and videos. However, encoding an image or a video as a sequence of patches results in costly attention patterns, as the requirements both in terms of memory and compute grow quadratically. To alleviate this problem, we propose a drop-in replacement for MHA called the Polynomial Mixer (PoM) that has the benefit of encoding the entire sequence into an explicit state. PoM has a linear complexity with respect to the number of tokens. This explicit state also allows us to generate frames in a sequential fashion, minimizing memory and compute requirement, while still being able to train in parallel. We show the Polynomial Mixer is a universal sequence-to-sequence approximator, just like regular MHA. We adapt several Diffusion Transformers (DiT) for generating images and videos with PoM replacing MHA, and we obtain high quality samples while using less computational resources. The code is available at https://github.com/davidpicard/HoMM.

Building hand-object models for dexterous in-hand manipulation remains a crucial and open problem. Major challenges include the difficulty of obtaining the geometric and dynamical models of the hand, object, and time-varying contacts, as well as the inevitable physical and perception uncertainties. Instead of building accurate models to map between the actuation inputs and the object motions, this work proposes to enable the hand-object systems to continuously approximate their local models via a self-identification process where an underlying manipulation model is estimated through a small number of exploratory actions and non-parametric learning. With a very small number of data points, as opposed to most data-driven methods, our system self-identifies the underlying manipulation models online through exploratory actions and non-parametric learning. By integrating the self-identified hand-object model into a model predictive control framework, the proposed system closes the control loop to provide high accuracy in-hand manipulation. Furthermore, the proposed self-identification is able to adaptively trigger online updates through additional exploratory actions, as soon as the self-identified local models render large discrepancies against the observed manipulation outcomes. We implemented the proposed approach on a sensorless underactuated Yale Model O hand with a single external camera to observe the object's motion. With extensive experiments, we show that the proposed self-identification approach can enable accurate and robust dexterous manipulation without requiring an accurate system model nor a large amount of data for offline training.

In applications that use knowledge representation (KR) techniques, in particular those that combine data-driven and logic methods, the domain of objects is not an abstract unstructured domain, but it exhibits a dedicated, deep structure of geometric objects. One example is the class of convex sets used to model natural concepts in conceptual spaces, which also links via convex optimization techniques to machine learning. In this paper we study logics for such geometric structures. Using the machinery of lattice theory, we describe an extension of minimal orthologic with a partial modularity rule that holds for closed convex cones. This logic combines a feasible data structure (exploiting convexity/conicity) with sufficient expressivity, including full orthonegation (exploiting conicity).