Dimensions underlying the representational alignment of deep neural networks with humans
2406.19087
0
0
Abstract
Determining the similarities and differences between humans and artificial intelligence is an important goal both in machine learning and cognitive neuroscience. However, similarities in representations only inform us about the degree of alignment, not the factors that determine it. Drawing upon recent developments in cognitive science, we propose a generic framework for yielding comparable representations in humans and deep neural networks (DNN). Applying this framework to humans and a DNN model of natural images revealed a low-dimensional DNN embedding of both visual and semantic dimensions. In contrast to humans, DNNs exhibited a clear dominance of visual over semantic features, indicating divergent strategies for representing images. While in-silico experiments showed seemingly-consistent interpretability of DNN dimensions, a direct comparison between human and DNN representations revealed substantial differences in how they process images. By making representations directly comparable, our results reveal important challenges for representational alignment, offering a means for improving their comparability.
Create account to get full access
Overview
- The paper explores the dimensions underlying the representational alignment between deep neural networks (DNNs) and human perception and cognition.
- It examines how the internal representations of DNNs reflect various conceptual and perceptual properties, and how these align with human understanding.
- The research provides insights into the cognitive and perceptual processes that shape the internal representations of DNNs.
Plain English Explanation
Deep neural networks (DNNs) are a type of artificial intelligence that can be trained to perform a wide range of tasks, from image recognition to language processing. These networks learn by processing large amounts of data and building internal representations of the information they're processing.
The researchers behind this paper were interested in understanding how the internal representations of DNNs align with the way humans perceive and conceptualize the world. They wanted to know which dimensions or aspects of the DNN representations correspond to the way humans think and experience their environment.
To do this, the researchers compared the internal representations of DNNs to measures of human perception and cognition. They found that the DNN representations reflect a diverse range of conceptual and perceptual properties, including things like [link: https://aimodels.fyi/papers/arxiv/saliency-suppressed-semantics-surfaced-visual-transformations-neural]visual salience[/link], [link: https://aimodels.fyi/papers/arxiv/post-hoc-manifold-explanations-analysis-facial-expression]facial expressions[/link], and [link: https://aimodels.fyi/papers/arxiv/learned-feature-representations-are-biased-by-complexity]complexity[/link].
The researchers also discovered that the DNN representations are shaped by factors like [link: https://aimodels.fyi/papers/arxiv/leveraging-human-ventral-visual-stream-to-improve]human visual processing[/link] and [link: https://aimodels.fyi/papers/arxiv/platonic-representation-hypothesis]conceptual categorization[/link]. This suggests that the internal representations of DNNs are not just mathematical abstractions, but are influenced by the same cognitive and perceptual processes that shape human understanding of the world.
Technical Explanation
The researchers used a combination of computational modeling, psychophysical experiments, and large-scale brain imaging data to investigate the dimensions underlying the representational alignment between DNNs and humans.
They first trained a DNN model on a large dataset of natural images and measured the internal representations of the model at different layers. They then compared these representations to a variety of human perceptual and cognitive measures, including visual salience, facial expression recognition, and conceptual categorization.
The results showed that the DNN representations reflected a diverse range of conceptual and perceptual properties. For example, the early layers of the DNN were found to align with measures of visual salience, while later layers aligned with more abstract, conceptual properties like facial expression recognition.
Further analyses revealed that the DNN representations were shaped by factors like human visual processing and conceptual categorization. This suggests that the internal representations of DNNs are not just mathematical abstractions, but are influenced by the same cognitive and perceptual processes that shape human understanding of the world.
Critical Analysis
The researchers acknowledge several limitations of their study. For example, they note that the DNN model they used was trained on a relatively narrow dataset of natural images, and that the alignment between DNN representations and human perception may be different for other types of data or tasks.
Additionally, the researchers caution that their findings do not imply that DNNs have human-like cognitive or perceptual capabilities. The alignment between DNN representations and human measures is likely due to the fact that both are shaped by similar underlying principles of information processing, rather than an indication that DNNs can think or perceive like humans.
Further research is needed to fully understand the relationship between DNN representations and human cognition. For example, it would be interesting to explore how the alignment between DNNs and humans changes as the models become more complex and are trained on more diverse datasets.
Overall, this study provides important insights into the cognitive and perceptual underpinnings of DNN representations, and highlights the potential for using computational models to better understand the human mind.
Conclusion
This paper presents an in-depth investigation into the dimensions underlying the representational alignment between deep neural networks (DNNs) and human perception and cognition. The researchers found that the internal representations of DNNs reflect a diverse range of conceptual and perceptual properties, including visual salience, facial expression recognition, and conceptual categorization.
Furthermore, the researchers discovered that these DNN representations are shaped by factors like human visual processing and conceptual categorization. This suggests that the internal representations of DNNs are not just mathematical abstractions, but are influenced by the same cognitive and perceptual processes that shape human understanding of the world.
These findings have important implications for our understanding of both artificial and human intelligence. By exploring the alignment between DNNs and humans, we can gain valuable insights into the cognitive and perceptual principles that underlie intelligent information processing in both biological and artificial systems.
This summary was produced with help from an AI and may contain inaccuracies - check out the links to read the original source documents!
Related Papers
New!Human-like object concept representations emerge naturally in multimodal large language models
Changde Du, Kaicheng Fu, Bincheng Wen, Yi Sun, Jie Peng, Wei Wei, Ying Gao, Shengpei Wang, Chuncheng Zhang, Jinpeng Li, Shuang Qiu, Le Chang, Huiguang He
0
0
The conceptualization and categorization of natural objects in the human mind have long intrigued cognitive scientists and neuroscientists, offering crucial insights into human perception and cognition. Recently, the rapid development of Large Language Models (LLMs) has raised the attractive question of whether these models can also develop human-like object representations through exposure to vast amounts of linguistic and multimodal data. In this study, we combined behavioral and neuroimaging analysis methods to uncover how the object concept representations in LLMs correlate with those of humans. By collecting large-scale datasets of 4.7 million triplet judgments from LLM and Multimodal LLM (MLLM), we were able to derive low-dimensional embeddings that capture the underlying similarity structure of 1,854 natural objects. The resulting 66-dimensional embeddings were found to be highly stable and predictive, and exhibited semantic clustering akin to human mental representations. Interestingly, the interpretability of the dimensions underlying these embeddings suggests that LLM and MLLM have developed human-like conceptual representations of natural objects. Further analysis demonstrated strong alignment between the identified model embeddings and neural activity patterns in many functionally defined brain ROIs (e.g., EBA, PPA, RSC and FFA). This provides compelling evidence that the object representations in LLMs, while not identical to those in the human, share fundamental commonalities that reflect key schemas of human conceptual knowledge. This study advances our understanding of machine intelligence and informs the development of more human-like artificial cognitive systems.
7/2/2024
Post-hoc and manifold explanations analysis of facial expression data based on deep learning
Yang Xiao
0
0
The complex information processing system of humans generates a lot of objective and subjective evaluations, making the exploration of human cognitive products of great cutting-edge theoretical value. In recent years, deep learning technologies, which are inspired by biological brain mechanisms, have made significant strides in the application of psychological or cognitive scientific research, particularly in the memorization and recognition of facial data. This paper investigates through experimental research how neural networks process and store facial expression data and associate these data with a range of psychological attributes produced by humans. Researchers utilized deep learning model VGG16, demonstrating that neural networks can learn and reproduce key features of facial data, thereby storing image memories. Moreover, the experimental results reveal the potential of deep learning models in understanding human emotions and cognitive processes and establish a manifold visualization interpretation of cognitive products or psychological attributes from a non-Euclidean space perspective, offering new insights into enhancing the explainability of AI. This study not only advances the application of AI technology in the field of psychology but also provides a new psychological theoretical understanding the information processing of the AI. The code is available in here: https://github.com/NKUShaw/Psychoinformatics.
4/30/2024
✨
Learned feature representations are biased by complexity, learning order, position, and more
Andrew Kyle Lampinen, Stephanie C. Y. Chan, Katherine Hermann
0
0
Representation learning, and interpreting learned representations, are key areas of focus in machine learning and neuroscience. Both fields generally use representations as a means to understand or improve a system's computations. In this work, however, we explore surprising dissociations between representation and computation that may pose challenges for such efforts. We create datasets in which we attempt to match the computational role that different features play, while manipulating other properties of the features or the data. We train various deep learning architectures to compute these multiple abstract features about their inputs. We find that their learned feature representations are systematically biased towards representing some features more strongly than others, depending upon extraneous properties such as feature complexity, the order in which features are learned, and the distribution of features over the inputs. For example, features that are simpler to compute or learned first tend to be represented more strongly and densely than features that are more complex or learned later, even if all features are learned equally well. We also explore how these biases are affected by architectures, optimizers, and training regimes (e.g., in transformers, features decoded earlier in the output sequence also tend to be represented more strongly). Our results help to characterize the inductive biases of gradient-based representation learning. These results also highlight a key challenge for interpretability $-$ or for comparing the representations of models and brains $-$ disentangling extraneous biases from the computationally important aspects of a system's internal representations.
6/7/2024
Saliency Suppressed, Semantics Surfaced: Visual Transformations in Neural Networks and the Brain
Gustaw Opie{l}ka, Jessica Loke, Steven Scholte
0
0
Deep learning algorithms lack human-interpretable accounts of how they transform raw visual input into a robust semantic understanding, which impedes comparisons between different architectures, training objectives, and the human brain. In this work, we take inspiration from neuroscience and employ representational approaches to shed light on how neural networks encode information at low (visual saliency) and high (semantic similarity) levels of abstraction. Moreover, we introduce a custom image dataset where we systematically manipulate salient and semantic information. We find that ResNets are more sensitive to saliency information than ViTs, when trained with object classification objectives. We uncover that networks suppress saliency in early layers, a process enhanced by natural language supervision (CLIP) in ResNets. CLIP also enhances semantic encoding in both architectures. Finally, we show that semantic encoding is a key factor in aligning AI with human visual perception, while saliency suppression is a non-brain-like strategy.
4/30/2024