Generalization

…Six debate motions examine generalization, counterfactual reasoning, data efficiency, evaluation, deployment costs and the limits of current benchmarks. Invited talks, contributed spotlights and posters connect theory to empirical comparisons, including negative results. Confirmed speakers include Chelsea Finn, Yunzhu Li, Georgia Chalvatzaki and Vincent Sitzmann. Paper submissions closed on 12 October; camera-ready papers are due on 6 November and

…theory or experiments on convergence, generalization, data freshness and differential privacy. The doctorate must be defended by 31 December 2026, and the appointment must provide a research environment distinct from the doctorate. The contract begins in France, with international visits encouraged. Appointment: 1 year, 4 months; advertised start 2027-01-01. 2 788 € gross monthly.

…supports both precise recollection and generalization. Her research combines activity-dependent cell tagging, interventions in neural circuits and analyses across the brain to investigate these transformations. The talk considers how changes in memory representations influence flexible behavior, including circumstances in which that flexibility becomes maladapti

…Yet many of their behaviours — generalization under overparametrization, grokking, OOD failures, neural scaling laws — recur across architectures and scales, and each, however surprising, can be reproduced and explained in controlled settings. I will review these efforts to identify and explain the universal phenomena of deep learning, and suggest that an LLM may amount to a sum of such tractable sub-phenomena, interpolative in nature. Finally, leaving scie

Deep networks display recurring phenomena such as grokking, scaling behaviour, generalization despite overparameterization and failures outside training distributions. This talk surveys controlled settings where those behaviours can be explained, suggesting that language models may combine many tractable, largely interpolative mechanisms. It then distinguishes that scientific account from a more speculative argument about the role of industrial computation and human labour, and what these imply for claims of broadly extrapolative intelligence.

Perfectly fitting the training data does not guarantee good generalization. For two idealized data distributions, this talk characterizes how frequently linear interpolating classifiers attain a given generalization error. Large-deviation and concentration results show why typical interpolators can perform worse than solutions reached by gradient descent and motivate the problem of sampling them.

…foundation models for biology, generalization beyond observed data, human evidence and scientific discovery. Hosted by Ravid Shwartz Ziv and Allen Roush. Watch the full conversation on the publisher’s YouTube channel.

…to understanding learning and generalization in brains and machines.

…these experiences, which supports generalization and prediction in similar situations in the future. One influential proposal for how the brain resolves this tension is that it separates the processes anatomically into Complementary Learning Systems, with the hippocampus rapidly encoding individual episodes and the neocortex slowly extracting regularities over days, months, and years. But this do

Equality reasoning is ubiquitous and purely abstract -- sameness or difference may be evaluated no matter the nature of the underlying objects. As a result, same-different tasks (SD) have been extensively studied as a starting point for understanding abstract reasoning in humans, and comparatively across animal species. With the rise of neural networks (NN) that exhibit a striking apparent proficiency for abstractions, equality reasoning in NNs has also gained interest. Yet despite the extensive study, conclusions about equality reasoning vary widely. Some authors claim that SD can be learned

Understanding how humans form knowledge structures from a stream of information is a question with implications for both human cognition and artificial intelligence. Structural learning is affected by the training regime, which can be either blocked into trials of similar forms or interleaved into trials of different forms. Previous studies show mixed results with some showing that blocked regimes improve different types of learning [1, 2, 3], while others show advantages of blocked regimes [4, 5, 6]. Additionally, few studies investigated the effect of regime on structural learning specifical

Abstraction and generalization are essential for flexible decision-making in novel situations. Recent work in humans and monkeys has shown how abstract variables are encoded by the representational geometry of single-neuron population activity in a format that supports generalization. However, these observations typically are made after learning has converged, leaving open the question of how these representations form. We designed the present study to tackle this question. We developed a factorized model of temporal abstraction that builds on the successor representation and is motivated by r

…However, the generalization of Visual Learning (VL) to unseen conditions varies across different tasks. Particularly, it has been shown that learning ‘easy’ visual tasks leads to better generalization than learning ‘hard’ ones. Moreover, experimental evidence suggests that greater generalizability in VL can be achieved through sequential curriculum training, specifically by training on ‘easier’ versions of a task before progressing to ‘harder’ versions. Yet, a neurocomputational explanation for

This study explores the fundamental aspects of acquired equivalence learning, where equivalence develops between seemingly unrelated stimuli due to shared consequences. The Rutgers Acquired Equivalence Test (RAET) is a paradigm developed to investigate visually guided equivalence learning, consisting of a learning phase and a test phase. The basal ganglia-frontal cortex pathways play a key role in the learning phase, and the test phase is linked to the functions of the hippocampus-medial temporal lobe.We developed three additional learning tests that differ in the complexity or modality of the

…Such generalization involves extracting common features from diverse experiences and utilizing them to make predictions in unfamiliar contexts. It remains unknown which brain areas are responsible for such abstractions. Neural representations in the medial entorhinal cortex (MEC) constitute a generalized map of the local space, but recent findings may suggest that the MEC possesses non-spatial dimensions suggesting a generalized network for learning state spaces irrespectiv

In the Rutgers Acquired Equivalence Test (RAET), participants are taught to associate pairs of visual stimuli (faces and fish) during the acquisition phase. Subsequently, in the test phase, they are required to recall and generalize these learned associations. We introduced a novel test (Polygon) that employs simpler visual stimuli with less semantic content, mirroring RAET's structure. This study was conducted to examine how the complexity and semantic content of visual stimuli impact the performance of migraine sufferers.The psychophysical responses of 41 migraine patients were evaluated acr

Developing scientifically-grounded theories for representation learning and generalization in artificial deep neural networks remains a grand challenge of fundamental interest to theoretical neuroscience and machine learning. I will discuss our work on one facet of this challenge — namely understanding generalization or “scaling laws” in learned neural networks as a function of basic control variables. I’ll discuss a taxonomy we develop that classifies different regimes of scaling behavior. We identify regimes where generalization exhibits universal scaling behavior and others where it can be

…these experiences, which supports generalization and prediction in similar situations in the future. One influential proposal for how the brain resolves this tension is that it separates the processes anatomically into Complementary Learning Systems, with the hippocampus rapidly encoding individual episodes and the neocortex slowly extracting regularities over days, months, and years. But this do

Humans and animals learn and plan with flexibility and efficiency well beyond that of modern Machine Learning methods. This is hypothesized to owe in part to the ability of animals to build structured representations of their environments, and modulate these representations to rapidly adapt to new settings. In the first part of this talk, I will discuss theoretical work describing how learned representations in hippocampus enable rapid adaptation to new goals by learning predictive representations. I will also cover work extending this account, in which we show how the predictive model can be

Humans and animals learn and plan with flexibility and efficiency well beyond that of modern Machine Learning methods. This is hypothesized to owe in part to the ability of animals to build structured representations of their environments, and modulate these representations to rapidly adapt to new settings. In the first part of this talk, I will discuss theoretical work describing how learned representations in hippocampus enable rapid adaptation to new goals by learning predictive representations, while entorhinal cortex compresses these predictive representations with spectral methods that s

Humans are natural teachers; learning through instruction is one of the most fundamental ways that we learn. Interactive Task Learning (ITL) is an emerging research agenda that studies the design of complex intelligent robots that can acquire new knowledge through natural human teacher-robot learner interactions. ITL methods are particularly useful for designing intelligent robots whose behavior can be adapted by humans collaborating with them. In this talk, I will summarize our recent findings on the structure that human instruction naturally has and motivate an intelligent system design that

Neuronal computation is mediated by spikes, yet it is unclear what the benefits of discrete spiking dynamics over continuous firing rates are. Many theoretical and computational studies treat single neurons as continuous units and their output as an effective firing rate. These works view spiking dynamics as a biophysical constraint, e.g., for energy efficiency. Conversely, other theories suggest that the exact timings of single spikes are meaningful. We propose a novel theory that shows the benefits of spiking dynamics on neural computation. In particular, we show that spiking neural dynamics

…Generalization and transfer learning in spatial domains is evident in mirror-invariant neural scene representations, and in the reuse of reference frames and representations across similar environments. Shared reference frames appear in other mammals as well, as, for example, in rodents reusing grid-cell maps for different but perceptually similar environments. However, the fie

…underexplored hypothesis suggests that generalization is facilitated by a low-dimensional organization of collective neural activity, yet evidence for the underlying neural mechanisms remains wanting. Combining