The Computational Memory Lab uses mathematical modeling and computational techniques to study human memory. We apply these quantitative methods both to data from laboratory studies of human memory and from electrophysiological studies done on patients with implanted electrodes.
Our current research is focused on basic mechanisms of episodic, spatial, and short-term recognition memory.
(Place location text below CML title) University of Pennsylvania, Department of Psychology Contact/Directions (make this a separate wiki page with the following info:)
3401 Walnut St., Suite 303C Philadelphia, PA 19104 Tel. 215-746-3500; Fax 215-746-6848 For directions to the lab, click here. (map image of our location instead of pdf, and revised directions that are inline)
Episodic memory refers to memory for events that are embedded in a temporal context. This includes both memory for significant life events and memory for common daily activities. In the laboratory, episodic memory is investigated by presenting lists of words for study, and then asking participants to recall the words. We have focused on the use of conditional probability and latency analysis (Kahana, M. J., 1996) to examine how participants transition from one recalled word to the next. These techniques quantify the order in which participants recall list items and the inter-response times between successive recalls (see Fig. 1).
Fig. 1: The contiguity effect in free recall. This curve shows the probability of making a recall to serial position i+lag immediately following recall of serial position i---that is, the conditional-response probability (CRP) as a function of lag.
Fig. 2: Brain oscillations associated with successful encoding are reinstated during correct retrieva. The top row of brain maps contrasts gamma-band oscillatory activity during the two second item presentation for items subsequently recalled and those that were forgotten. The bottom row contrasts gamma-band oscillations during the 500 milliseconds preceding recall verbalization for correct items and for prior-list intrusions. In each map, red corresponds to regions where the contrast was significant, gray to non-significant contrasts, and black indicates brain regions excluded from the analysis due to insufficient electrode coverage.
To explain the recency and contiguity effects in free recall, Howard and Kahana (2002) developed the Temporal Context Model of episodic memory. TCM is a distributed memory model that specifies the mechanisms of contextual drift and contextual retrieval. Through the drift mechanism, TCM describes how a temporal code is created by the integration of recently retrieved contextual states. As such, TCM represents the first formal model of how memories become 'episodic' (linked to the time when they occurred). TCM also provides an alternative explanation for associative tendencies in recall. Rather than resulting from co-occurrence in short-term memory (the standard earlier view), TCM suggests that these tendencies appear because recall of an item recovers the temporal context for the item, which in turn cues recall of subsequent items. Similarly, recency effects appear because the temporal context at the time of the memory test is most similar to the temporal context associated with recent items. Unlike short-term memory based models, TCM predicts that recency and associative effects should be approximately time-scale invariant (Howard, M. W. and Kahana, M. J., 1999, Sederberg, et al., 2008).
In addition to behavioral and theoretical analyses of episodic memory, we also explore the neurophysiology of episodic memory with both scalp and intracranial electroencephalographic (iEEG) recordings. Intracranial recordings can be obtained from epilepsy patients who have had electrodes surgically implanted on the cortical surface of the brain or through the medial temporal lobes (including hippocampus) as part of the clinical process of localizing seizure foci. One focus of this research is to determine the oscillatory correlates of successful episodic memory formation. Our lab has found that 44-100 Hz (gamma) brain oscillations increase while participants are studying words that they will successfully, as opposed to unsuccessfully, recall (Sederberg, et al., 2006). The same distribution of gamma activity across both hippocampus and neocortex is reactivated just prior to recalling an item, with higher levels of gamma predicting whether or not the recalled item was actually studied (Sederberg, et al., 2007; see Fig. 2).
To study human spatial cognition, we developed a virtual reality taxi-driver game (Fig. 1), which encourages participants to find efficient paths between arbitrary spatial locations and landmarks. To download a sample of a YellowCab session, click here.
In a series of studies, we have documented the existence and character of the theta rhythm in the human brain as participants learned to navigate through complex virtual environments (Kahana, et al., 1999; Caplan, et al., 2001; Caplan, et al., 2003; Ekstrom, et al., 2005; Jacobs, et al., Submitted).
By recording the behavior of individual brain cells (in collaboration with UCLA Neurosurgeon Dr. Itzhak Fried), we have identified place cells in the human brain. These cells, which are found primarily in the human hippocampus, become highly active when a given spatial location is being traversed from any direction. We also identified two other cellular responses in the human brain: cells that become active in response to viewing a salient landmark (from any location) and cells that become active when searching for a particular goal location (irrespective of location or view). Finally, we found a large number of cells that represent combinations of these three features. These results were recently reported by Ekstrom, et al. (2003).
Fig. 1: Example of the YellowCab task. We examined whether the two key physiological markers of spatial navigation in rodents might have parallels in the human brain. When rodents navigate through a novel environment recordings of electrical activity from the hippocampus (and nearby brain structures) reveal a striking 4-10 Hz rhythmic oscillation known as the hippocampal theta rhythm. At the same time, certain cells in the hippocampus, termed place cells, increase their rate of activity when particular regions of the space are being traversed. These two phenomena figure prominently in animal models of learning and spatial navigation.
Fig. 2: Firing-rate map of a right hippocampal cell showing significant place selectivity. Lettered squares (SA,SB,SC) indicate target store locations, white boxes indicate non-target buildings, red lines indicate the subject's trajectory, and the red square indicates regions of significantly high firing rate (all examples, p < 0.01).
Fig. 3: Anatomical distribution of place cells. Place-responsive cells were clustered in the hippocampus (H) compared with amygdala (A), parahippocampal region (PR) and frontal lobes (FR).
Our lab has also worked on developing and testing the predictions of quantitative models of recognition memory. We have developed, together with Robert Sekuler, a mathematical model of recognition memory that goes beyond existing models in predicting the accuracy of participants' responses on individual lists (Kahana, M. J. and Sekuler, R., 2002). Our noisy exemplar model (NEMO) is based on the idea that judging whether an item was on a recently studied list depends not only on the similarity of the item to the items in the list (as in other standard models of recognition memory), but also on the similarities among the items within the studied list. NEMO is thus capturing a context effect in the data--the similarity of a study item to the items in the list is judged relative to the overall level of similarity among the list items. If one has just studied a list of highly dissimilar items, one would tend to judge a moderately similar item as having been on the studied list; if one has just studied a list of highly similar items, it will be relatively easy to reject the moderately similar item as not having been on the list.
We then wondered how this variation in difficulty due to similarity, which we called similarity-based interference. In this form of interference, you confuse the probe with similar items in the list you had to study. Similarity-based interference was correlated with oscillatory activity in the brain. In a scalp EEG study (van Vugt et al., submitted) we found that similarity-based interference was associated with the amplitude (power) of oscillations in the theta band (4-8 Hz) at the time the probe is shown. This similarity-based interference had a quite different neural signature from proactive interference. In trials with high proactive interference, you confuse the probe because you have recently seen it and cannot remember what list it was on, and this was associated with occipital gamma oscillations.