How does experience inform decisions? In episodic sampling, decisions are guided by a few episodic memories of past choices. This process can yield choice patterns similar to model-free reinforcement learning; however, samples can vary from trial to trial, causing decisions to vary. Here we show that context retrieved during episodic sampling can cause choice behavior to deviate sharply from the predictions of reinforcement learning. Specifically, we show that, when a given memory is sampled, choices (in the present) are influenced by the properties of other decisions made in the same context as the sampled event. This effect is mediated by fMRI measures of context retrieval on each trial, suggesting a mechanism whereby cues trigger retrieval of context, which then triggers retrieval of other decisions from that context. This result establishes a new avenue by which experience can guide choice and, as such, has broad implications for the study of decisions.

Generalization to new tasks requires learning of task representations that accurately reflect the similarity structure of the task space. Here, we argue that episodic memory (EM) plays an essential role in this process by stabilizing task representations, thereby supporting the accumulation of structured knowledge. We demonstrate this using a neural network model that infers task representations that minimize the current task's objective function; crucially, the model can retrieve previously encoded task representations from EM and use these to initialize the task inference process. With EM, the model succeeds in learning the underlying task structure; without EM, task representations drift and the network fails to learn the structure. We further show that EM errors can support structure learning by promoting the activation of similar task representations in tasks with similar sensory inputs. Overall, this model provides a novel account of how EM supports the acquisition of structured task representations.

How are exceptions to a generalization learned? 20 participants were exposed to a mini-artificial language in whicheach word (prefix + wordstem) was associated with a unique image. One of two prefixes generalized probabilistically:it appeared with 40 stems associated with faces and 8 exceptions, which were associated with scenes. The other prefixoccurred with 8 faces and 8 scenes. The prefixes and the image categories (faces vs. scenes) were counterbalanced acrossparticipants. Participants performed a 2 alternative-forced choice task on all items, with feedback, over 6 repeating blocks.Results show that image-word pairs that included the generalizable prefix were learned better than those which appearedwith the other prefix, despite having 48 items in the first class and 16 in the other (d = 1.28, d = 0.80, p = 0.023). Weinvestigate the neural representation of these words and how they change over the course of learning.

The method of loci is a powerful mnemonic technique for memorizing a list of unrelated items. With a pre-specified routein a familiar ”memory palace”, one can encode material by attaching items to loci along this route, and later effectivelyrecall them by mentally walking along the same route. Despite its efficacy, there is no existing model that explains whythe method of loci promotes memory improvement during memory search. To fill this gap, we provide a rational accountof why the method of loci improves memory. We define memory search as a task with the goal of minimizing retrievalcost, and demonstrate that the method of loci gives an optimal policy for this task. We discuss the implications of thisresult, and compare it with models of memory search without using mnemonic techniques.

We remember when things change. Particularly salient are experiences where there is a change in rewards, eliciting reward prediction errors (RPEs). How do RPEs influence our memory of those experiences? One idea is that this signal directly enhances the encoding of memory. Another, not mutually exclusive, idea is that the RPE signals a deeper change in the environment, leading to the mnemonic separation of subsequent experiences from what came before, thereby creating a new latent context and a more separate memory trace. We tested this in four experiments where participants learned to predict rewards associated with a series of trial-unique images. High-magnitude RPEs indicated a change in the underlying distribution of rewards. To test whether these large RPEs created a new latent context, we first assessed recognition priming for sequential pairs that included a high-RPE event or not (Exp. 1: n = 27 & Exp. 2: n = 83). We found evidence of recognition priming for the high-RPE event, indicating that the high-RPE event is bound to its predecessor in memory. Given that high-RPE events are themselves preferentially remembered (Rouhani, Norman, & Niv, 2018), we next tested whether there was an event boundary across a high-RPE event (i.e., excluding the high-RPE event itself; Exp. 3: n = 85). Here, sequential pairs across a high RPE no longer showed recognition priming whereas pairs within the same latent reward state did, providing initial evidence for an RPE-modulated event boundary. We then investigated whether RPE event boundaries disrupt temporal memory by asking participants to order and estimate the distance between two events that had either included a high-RPE event between them or not (Exp. 4). We found (n = 49) and replicated (n = 77) worse sequence memory for events across a high RPE. In line with our recognition priming results, we did not find sequence memory to be impaired between the high-RPE event and its predecessor, but instead found worse sequence memory for pairs across a high-RPE event. Moreover, greater distance between events at encoding led to better sequence memory for events across a low-RPE event, but not a high-RPE event, suggesting separate mechanisms for the temporal ordering of events within versus across a latent reward context. Altogether, these findings demonstrate that high-RPE events are both more strongly encoded, show intact links with their predecessor, and act as event boundaries that interrupt the sequential integration of events. We captured these effects in a variant of the Context Maintenance and Retrieval model (CMR; Polyn, Norman, & Kahana, 2009), modified to incorporate RPEs into the encoding process.

How are exceptions to a probabilistic generalization learned? The present results suggest exceptions are learned in partby selectively suppressing the competing category, as opposed to only increasing knowledge of exceptions. Participantswere exposed to a mini-artificial language with a probabilistic generalization (80-20%) that mapped labels to categories ofimages (faces and scenes). Mouse-tracking trajectories determined the degree to which the generalization served as a lureto exceptions, compared to a separate baseline condition. Over time, the generalization became suppressed in a context-sensitive way: for exception items only. This extends retrieval induced forgetting, in which a particular item is suppresseddue to competition from partial retrieval, to include the entire conceptual category. Post-test revealed high item-specificaccuracy, even though category recognition was sufficient for the task.

There has been a great deal of interest in howgeneralizations and exceptions are learned, withparticular interest in how speakers learn to avoidovergeneralizations. Do overgeneralizations disappearonly because exceptions become more stronglyrepresented or does the generalization itself becomesuppressed? Novel labels were constructed bycombining 56 syllables with one of two prefixes, andeach label was assigned a unique image. Most labels withthe first prefix were paired with images from ageneralization category, whereas exceptional labels werepaired with images from a different semantic category.All labels with the second prefix appeared with a thirdcategory (“baseline”). Participants used a computermouse to choose one of two images for each label.Mouse-tracking results show that the generalizationitself became suppressed over time in the context ofexceptional labels. A post-test demonstrated thatexceptions were learned with item-specific precision.

Shared structure in neural responses across people can be obscured because these neural responses sit on different ”co-ordinate systems”; hyperalignment can recover this shared structure by placing different people’s brain responses intoa common functional space (Chen et al., 2015; Haxby et al., 2011). Here, we apply this framework to understand thehidden representations of neural networks. Different neural networks can represent the same input-output mapping usingvery different weights. We show that hyperalignment can construct a shared representational space that recovers sharedrepresentation structure across neural networks. We formally connect representational similarity analysis and hyperalign-ment and use simulations to demonstrate the robustness of hyperalignment against several types of transformations thatpreserve the representation geometry of the network. We also empirically tested our method on some supervised learningbenchmarks (CIFAR10, MNIST) for both standard and convolutional networks.

A fundamental question in memory research is how different forms of memory interact. Previous research has shown that people rely on working memory (WM) in short-term recognition tasks; a common view is that episodic memory (EM) only influences performance on these tasks when WM maintenance is disrupted. However, retrieval of memories from EM has been widely observed during brief periods of quiescence, raising the possibility that EM retrievals during maintenance-critically, before a response can be prepared-might affect short-term recognition memory performance even in the absence of distraction. We hypothesized that this influence would be mediated by the lingering presence of reactivated EM content in WM. We obtained support for this hypothesis in three experiments, showing that delay-period EM reactivation introduces incidentally-associated information ( context ) into WM, and that these retrieved associations negatively impact subsequent recognition, leading to substitution errors (Experiment 1) and slowing of accurate responses (Experiment 2). fMRI pattern analysis showed that slowing is mediated by the content of EM reinstatement (Experiment 3). These results expose a previously hidden influence of EM on WM, raising new questions about the adaptive nature of their interaction.

Event segmentation theory posits that we segment continuous experience into discrete events and event boundaries occur at large transient increases in prediction error, often related to context changes. Identifying event boundaries a priori has been difficult in naturalistic settings. To overcome this challenge for story listening, we used a deep learning language model (GPT2) to compute the predicted probability distributions of the next word, at each point in the story. For 3 stories, we computed the surprise, the entropy, and the Kullback-Leibler divergence (KLD). We then asked participants to listen to these stories while marking event boundaries. We used regression models to compare the GPT2 measures and the human segmentation data. Preliminary results indicate that event boundaries are associated with transient increases in KLD. This supports the hypothesis that prediction error serves as a control mechanism governing event segmentation, and points to important differences between operational definitions of prediction error.