We propose firefly neural architecture descent, a general framework for progressively and dynamically growing neural networks to jointly optimize the networks' parameters and architectures. Our method works in a steepest descent fashion, which iteratively finds the best network within a functional neighborhood of the original network that includes a diverse set of candidate network structures. By using Taylor approximation, the optimal network structure in the neighborhood can be found with a greedy selection procedure. We show that firefly descent can flexibly grow networks both wider and deeper, and can be applied to learn accurate but resource-efficient neural architectures that avoid catastrophic forgetting in continual learning. Empirically, firefly descent achieves promising results on both neural architecture search and continual learning.

To reduce the human efforts in neural network design, Neural Architecture Search (NAS) has been applied with remarkable success to various high-level vision tasks such as classification and semantic segmentation. The underlying idea for the NAS algorithm is straightforward, namely, to allow the network the ability to choose among a set of operations (\eg convolution with different filter sizes), one is able to find an optimal architecture that is better adapted to the problem at hand. However, so far the success of NAS has not been enjoyed by low-level geometric vision tasks such as stereo matching. This is partly due to the fact that state-of-the-art deep stereo matching networks, designed by humans, are already sheer in size. Directly applying the NAS to such massive structures is computationally prohibitive based on the currently available mainstream computing resources. In this paper, we propose the first \emph{end-to-end} hierarchical NAS framework for deep stereo matching by incorporating task-specific human knowledge into the neural architecture search framework.

Neural architecture search (NAS) has been extensively studied in the past few years. A popular approach is to represent each neural architecture in the search space as a directed acyclic graph (DAG), and then search over all DAGs by encoding the adjacency matrix and list of operations as a set of hyperparameters. Recent work has demonstrated that even small changes to the way each architecture is encoded can have a significant effect on the performance of NAS algorithms. In this work, we present the first formal study on the effect of architecture encodings for NAS, including a theoretical grounding and an empirical study. First we formally define architecture encodings and give a theoretical characterization on the scalability of the encodings we study.

Denoising Diffusion Models (DDMs) are widely used for high-quality image generation and medical image segmentation but often rely on Unet-based architectures, leading to high computational overhead, especially with high-resolution images. This work proposes three NCA-based improvements for diffusion-based medical image segmentation. First, Multi-MedSegDiffNCA uses a multilevel NCA framework to refine rough noise estimates generated by lower level NCA models. Second, CBAM-MedSegDiffNCA incorporates channel and spatial attention for improved segmentation. Third, MultiCBAM-MedSegDiffNCA combines these methods with a new RGB channel loss for semantic guidance. Evaluations on Lesion segmentation show that MultiCBAM-MedSegDiffNCA matches Unet-based model performance with dice score of 87.84% while using 60-110 times fewer parameters, offering a more efficient solution for low resource medical settings.

In the fields of computation and neuroscience, much is still unknown about the underlying computations that enable key cognitive functions including learning, memory, abstraction and behavior. This paper proposes a mathematical and computational model of learning and memory based on a small set of bio-plausible functions that include coincidence detection, signal modulation, and reward/penalty mechanisms. Our theoretical approach proposes that these basic functions are sufficient to establish and modulate an information space over which computation can be carried out, generating signal gradients usable for inference and behavior. The computational method used to test this is a structurally dynamic cellular automaton with continuous-valued cell states and a series of recursive steps propagating over an undirected graph with the memory function embedded entirely in the creation and modulation of graph edges. The experimental results show: that the toy model can make near-optimal choices to re-discover a reward state after a single training run; that it can avoid complex penalty configurations; that signal modulation and network plasticity can generate exploratory behaviors in sparse reward environments; that the model generates context-dependent memory representations; and that it exhibits high computational efficiency because of its minimal, single-pass training requirements combined with flexible and contextual memory representation.

The primary promise of decentralized learning is to allow users to engage in the training of machine learning models in a collaborative manner while keeping their data on their premises and without relying on any central entity. However, this paradigm necessitates the exchange of model parameters or gradients between peers. Such exchanges can be exploited to infer sensitive information about training data, which is achieved through privacy attacks (e.g Membership Inference Attacks -- MIA). In order to devise effective defense mechanisms, it is important to understand the factors that increase/reduce the vulnerability of a given decentralized learning architecture to MIA. In this study, we extensively explore the vulnerability to MIA of various decentralized learning architectures by varying the graph structure (e.g number of neighbors), the graph dynamics, and the aggregation strategy, across diverse datasets and data distributions. Our key finding, which to the best of our knowledge we are the first to report, is that the vulnerability to MIA is heavily correlated to (i) the local model mixing strategy performed by each node upon reception of models from neighboring nodes and (ii) the global mixing properties of the communication graph. We illustrate these results experimentally using four datasets and by theoretically analyzing the mixing properties of various decentralized architectures. Our paper draws a set of lessons learned for devising decentralized learning systems that reduce by design the vulnerability to MIA.

GNAS (Graph Neural Architecture Search) has demonstrated great effectiveness in automatically designing the optimal graph neural architectures for multiple downstream tasks, such as node classification and link prediction. However, most existing GNAS methods cannot efficiently handle large-scale graphs containing more than million-scale nodes and edges due to the expensive computational and memory overhead. To scale GNAS on large graphs while achieving better performance, we propose SA-GNAS, a novel framework based on seed architecture expansion for efficient large-scale GNAS. Similar to the cell expansion in biotechnology, we first construct a seed architecture and then expand the seed architecture iteratively. Specifically, we first propose a performance ranking consistency-based seed architecture selection method, which selects the architecture searched on the subgraph that best matches the original large-scale graph. Then, we propose an entropy minimization-based seed architecture expansion method to further improve the performance of the seed architecture. Extensive experimental results on five large-scale graphs demonstrate that the proposed SA-GNAS outperforms human-designed state-of-the-art GNN architectures and existing graph NAS methods. Moreover, SA-GNAS can significantly reduce the search time, showing better search efficiency. For the largest graph with billion edges, SA-GNAS can achieve 2.8 times speedup compared to the SOTA large-scale GNAS method GAUSS. Additionally, since SA-GNAS is inherently parallelized, the search efficiency can be further improved with more GPUs. SA-GNAS is available at https://github.com/PasaLab/SAGNAS.

Large Language Models demonstrate remarkable mathematical capabilities but at the same time struggle with abstract reasoning and planning. In this study, we explore whether Transformers can learn to abstract and generalize the rules governing Elementary Cellular Automata. By training Transformers on state sequences generated with random initial conditions and local rules, we show that they can generalize across different Boolean functions of fixed arity, effectively abstracting the underlying rules. While the models achieve high accuracy in next-state prediction, their performance declines sharply in multi-step planning tasks without intermediate context. Our analysis reveals that including future states or rule prediction in the training loss enhances the models' ability to form internal representations of the rules, leading to improved performance in longer planning horizons and autoregressive generation. Furthermore, we confirm that increasing the model's depth plays a crucial role in extended sequential computations required for complex reasoning tasks. This highlights the potential to improve LLM with inclusion of longer horizons in loss function, as well as incorporating recurrence and adaptive computation time for dynamic control of model depth.

Knowledge tracing (KT) aims to trace students' knowledge states by predicting whether students answer correctly on exercises. Despite the excellent performance of existing Transformer-based KT approaches, they are criticized for the manually selected input features for fusion and the defect of single global context modelling to directly capture students' forgetting behavior in KT, when the related records are distant from the current record in terms of time. To address the issues, this paper first considers adding convolution operations to the Transformer to enhance its local context modelling ability used for students' forgetting behavior, then proposes an evolutionary neural architecture search approach to automate the input feature selection and automatically determine where to apply which operation for achieving the balancing of the local/global context modelling. In the search space, the original global path containing the attention module in Transformer is replaced with the sum of a global path and a local path that could contain different convolutions, and the selection of input features is also considered. To search the best architecture, we employ an effective evolutionary algorithm to explore the search space and also suggest a search space reduction strategy to accelerate the convergence of the algorithm.

One-shot weight sharing methods have recently drawn great attention in neural architecture search due to high efficiency and competitive performance. However, weight sharing across models has an inherent deficiency, i.e., insufficient training of subnetworks in the hypernetwork. To alleviate this problem, we present a simple yet effective architecture distillation method. The central idea is that subnetworks can learn collaboratively and teach each other throughout the training process, aiming to boost the convergence of individual models. We introduce the concept of prioritized path, which refers to the architecture candidates exhibiting superior performance during training.