Inner interpretability is a promising field aiming to uncover the internal mechanisms of AI systems through scalable, automated methods. While significant research has been conducted on large language models, limited attention has been paid to applying inner interpretability to large-scale image tasks, focusing primarily on architectural and functional levels to visualize learned concepts. In this paper, we first present a conceptual framework that supports inner interpretability and multilevel analysis for large-scale image classification tasks. Specifically, we introduce the Bi-directional Interaction between Concept and Input Embeddings (Bi-ICE) module, which facilitates interpretability across the computational, algorithmic, and implementation levels. This module enhances transparency by generating predictions based on human-understandable concepts, quantifying their contributions, and localizing them within the inputs. Finally, we showcase enhanced transparency in image classification, measuring concept contributions, and pinpointing their locations within the inputs. Our approach highlights algorithmic interpretability by demonstrating the process of concept learning and its convergence.

Wearable sensors have found numerous applications in health and wellness promotion and have achieved great success leveraging advancements in deep learning. However, the development of robust continues to be hindered by issues related to sensor noise, inconsistent sampling rates, and individual differences. Topological data analysis (TDA) has emerged as a viable solution to extract robust features from such time-series data by converting them into persistence images (PIs), which capture intrinsic characteristics and demonstrate resilience to noise and signal variations. However, the computational costs of TDA pose significant challenges for small devices with limited resources. To more efficiently incorporate topological features, we utilize knowledge distillation (KD), which is a promising way to generate a smaller model using larger models. Multiple teachers can be adopted to enrich features in KD. However, this approach has presented two key challenges: 1) differences in feature dimensions from multimodal data and 2) conflicting knowledge provided by the different teachers, both of which can degrade the student model's performance. To address these issues, we propose a novel KD framework called multimodal global latent workspace-based KD (mGLW-KD) that is motivated by global workspace theory (GTW) from cognitive neuroscience. GWT models how the brain integrates and distributes relevant information across different neural modules through a shared workspace, and it includes attentional control and working memory to prioritize and retain key information. Inspired by this theory, mGLW-KD incorporates a working memory module to unify diverse knowledge from multiple teacher models into a shared latent workspace, facilitating efficient knowledge transfer to the student model. By integrating topological insights with cognitive principles, mGLW-KD addresses the challenges posed by wearable sensor data and enables the student model to achieve superior performance using only time-series input during inference.

Many interpretable AI approaches have been proposed to provide plausible explanations for a model’s decision-making. However, configuring an explainable model that effectively communicates among computational modules has received less attention. A recently proposed shared global workspace theory showed that networks of distributed modules can benefit from sharing information with a bottle-necked memory because the communication constraints encourage specialization, compositionality, and synchronization among the modules. Inspired by this, we propose Concept-Centric Transformers, a simple yet effective configuration of the shared global workspace for interpretability, consisting of: i) an object-centric-based memory module for extracting semantic concepts from input features, ii) a cross-attention mechanism between the learned concept and input embeddings, and iii) standard classification and explanation losses to allow human analysts to directly assess an explanation for the model’s classification reasoning. We test our approach against other existing concept-based methods on classification tasks for various datasets, including CIFAR100, CUB-200-2011, and ImageNet, and we show that our model achieves better classification accuracy than all baselines across all problems but also generates more consistent concept-based explanations of classification output.

Biology is both an important application area and a source of motivation for development of advanced machine learning techniques. Although much attention has been paid to large and complex data sets resulting from high-throughput sequencing, advances in high-quality video recording technology have begun to generate similarly rich data sets requiring sophisticated techniques from both computer vision and time-series analysis. Moreover, just as studying gene expression patterns in one organism can reveal general principles that apply to other organisms, the study of complex social interactions in an experimentally tractable model system, such as a laboratory ant colony, can provide general principles about the dynamics of other social groups. Here, we focus on one such example from the study of reproductive regulation in small laboratory colonies of more than 50 Harpegnathos ants. These ants can be artificially induced to begin a ~20 day process of hierarchy reformation. Although the conclusion of this process is conspicuous to a human observer, it remains unclear which behaviors during the transient period are contributing to the process. To address this issue, we explore the potential application of One-class Classification (OC) to the detection of abnormal states in ant colonies for which behavioral data is only available for the normal societal conditions during training. Specifically, we build upon the Deep Support Vector Data Description (DSVDD) and introduce the Inner-Outlier Generator (IO-GEN) that synthesizes fake “inner outlier” observations during training that are near the center of the DSVDD data description.We show that IO-GEN increases the reliability of the final OC classifier relative to other DSVDD baselines. This method can be used to screen video frames for which additional human observation is needed. Although we focus on an application with laboratory colonies of social insects, this approach may be applied to video data from other social systems to either better understand the causal factors behind social phase transitions or even to predict the onset of future transitions.

When members of a multi-robot team follow regular motion rules sensitive to robots and other environmental factors within sensing range, the team itself may become an informational fabric for gaining situational awareness without explicit signalling among robots. In our previous work [1], we used machine learning to develop a scalable module, trained only on data from 3-robot teams, that could predict the positions of all robots in larger multi-robot teams based only on observations of the movement of a robot's nearest neighbor. Not only was this approach scalable from 3-to-many robots, but it did not require knowledge of the control laws of the robots under observation, as would a traditional observer-based approach. However, performance was only tested in simulation and could only be a substitute for explicit communication for short periods of time or in cases of very low sensing noise. In this work, we apply more sophisticated machine learning methods to data from a physically realized robotic team to develop Remote Teammate Localization (ReTLo) modules that can be used in realistic environments. To be specific, we adopt Long-Short-Term-Memory (LSTM) [2] to learn the evolution of behaviors in a modular team, which has the effect of greatly reducing errors from regression outcomes. In contrast with our previous work in simulation, all of the experiments conducted in this work were conducted on the Thymio physical, two-wheeled robotic platform.

Cellular automata have been useful artificial models for exploring how relatively simple rules combined with spatial memory can give rise to complex emergent patterns. Moreover, studying the dynamics of how rules emerge under artificial selection for function has recently become a powerful tool for understanding how evolution can innovate within its genetic rule space. However, conventional cellular automata lack the kind of state feedback that is surely present in natural evolving systems. Each new generation of a population leaves an indelible mark on its environment and thus affects the selective pressures that shape future generations of that population. To model this phenomenon, we have augmented traditional cellular automata with state-dependent feedback. Rather than generating automata executions from an initial condition and a static rule, we introduce mappings which generate iteration rules from the cellular automaton itself. We show that these new automata contain disconnected regions which locally act like conventional automata, thus encapsulating multiple functions into one structure. Consequently, we have provided a new model for processes like cell differentiation. Finally, by studying the size of these regions, we provide additional evidence that the dynamics of self-reference may be critical to understanding the evolution of natural language. In particular, the rules of elementary cellular automata appear to be distributed in the same way as words in the corpus of a natural language.

This paper introduces a novel framework for the design of distributed agents that must complete externally generated tasks but also can volunteer to process tasks encountered by other agents. To reduce the computational and communication burden of coordination between agents to perfectly balance load around the network, the agents adjust their volunteering propensity asynchronously within a fictitious trading economy. This economy provides incentives for nontrivial levels of volunteering for remote tasks, and thus load is shared. Moreover, the combined effects of diminishing marginal returns and network topology lead to competitive equilibria that have task reallocations that are qualitatively similar to what is expected in a load-balancing system with explicit coordination between nodes. In the paper, topological and algorithmic conditions are given that ensure the existence and uniqueness of a competitive equilibrium. Additionally, a decentralized distributed gradient-ascent algorithm is given that is guaranteed to converge to this equilibrium while not causing any node to over-volunteer beyond its maximum task-processing rate. The framework is applied to an autonomous-air-vehicle example, and connections are drawn to classic studies of the evolution of cooperation in nature.

Collective food transport in ant colonies is a striking, albeit poorly understood, example of coordinated group behavior in nature that can serve as a template for robust, decentralized multi-robot cooperative manipulation strategies. We investigate this behavior in Aphaenogaster cockerelli ants in order to derive a model of the ants' roles and behavioral transitions and the resulting dynamics of a transported load. In experimental trials, A. cockerelli are induced to transport a rigid artificial load to their nest. From video recordings of the trials, we obtain time series data on the load position and the population counts of ants in three roles. From our observations, we develop a stochastic hybrid system model that describes the time evolution of these variables and that can be used to derive the dynamics of their statistical moments. In our model, ants switch stochastically between roles at constant, unknown probability rates, and ants in one role pull on the load with a force that acts as a proportional controller on the load velocity with unknown gain and set point. We compute these unknown parameters by using standard numerical optimization techniques to fit the time evolution of the means of the load position and population counts to the averaged experimental time series. The close fit of our model to the averaged data and to data for individual trials demonstrates the accuracy of our proposed model in predicting the ant behavior.