Lecture 12 Todd M. Gureckis, Lila Davachi

Transcription

1 An Introduction to Learning Lecture 12 Todd M. Gureckis, Lila Davachi Department of Psychology New York University 1

2 Interference: Brown-Peterson Paradigm - simple paradigm for looking at interference effects - present triplets of stimuli - have people count backwards for some interval - recall the most recently presented triplet

3 Interference: Brown-Peterson Paradigm - 3 words will appear (e.g., apple orange banana) - then you will see a 4-digit number (e.g., 2764) - when the 4 digit number appears, start counting down by 7s from that number until I say stop

4 Brown-Peterson Demo Pontiac Lexus Mazda

5 Brown-Peterson Demo 1721

6 Brown-Peterson Demo Pontiac Lexus Mazda

7 Brown-Peterson Demo Subaru Honda Volvo

8 Brown-Peterson Demo 9853

9 Brown-Peterson Demo Subaru Honda Volvo

10 Brown-Peterson Demo Isuzu Buick Nissan

11 Brown-Peterson Demo 1533

12 Brown-Peterson Demo Isuzu Buick Nissan

13 Brown-Peterson Demo Volkswagen Pontiac Mitsubishi

14 Brown-Peterson Demo 2206

15 Brown-Peterson Demo Volkswagen Pontiac Mitsubishi

16 Brown-Peterson Demo Acura Ferrari Oldsmobile

17 Brown-Peterson Demo 5428

18 Brown-Peterson Demo Acura Ferrari Oldsmobile

19 Brown-Peterson Demo Chevrolet Saab Cadillac

20 Brown-Peterson Demo 6612

21 Brown-Peterson Demo Chevrolet Saab Cadillac

22 Brown-Peterson Demo sparrow eagle bluejay

23 Brown-Peterson Demo 3486

24 Brown-Peterson Demo sparrow eagle bluejay

25 Important BP Findings - Improved performance with category switch - Suggests that proactive interference is contributing to forgetting - Some studies have looked at within-subject delay manipulations: short delay vs long delay Increasing delay between study & test hurts performance Evidence for decay...??? However, no effect of delay on the first trial Suggests that decay is not occurring

26 Important BP Findings If decay isn t contributing to forgetting, then why does increasing the retention interval hurt performance (on trials after the first one)? Put another way: How can we get time-dependent forgetting in the absence of decay?

27 Contextual Drift Switch at 96th St itchy Work hungry 9:40 9:41 9:42 9:43 The constellation of thoughts in our heads drifts over time... Thanks to K.Norman for slides

28 Contextual Drift Switch at 96th St itchy Work hungry printer 9:40 9:41 9:42 9:43 Items presented at a particular time get associated with active contextual elements If you study printer at 9:42, it gets associated with hungry, work, and switch at 96th St

29 Contextual Drift Switch at 96th St itchy Work hungry printer test 9:40 9:41 9:42 9:43 If you want to figure out what you studied recently, cue memory with the current set of contextual elements If, at 9:43, you want to recall recently studied items, cue with hungry and switch at 96th St

30 Contextual Drift Switch at 96th St itchy Work hungry printer test 9:40 9:41 9:42 9:43 Because (earlier) you associated printer with hungry and switch at 96th St, cuing with hungry and switch at 96th St should trigger recall of printer

31 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 S2 In the Brown-Peterson test, we have multiple study and test phases (S1, T1, S2, T2, etc.) Each study list gets associated with the currently active contextual elements e.g., S1 gets associated with itchy, Work, and switch at 96th St T2

32 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 S2 T2 9:40 9:41 9:42 9:43 Goal: you want to recall the most recent study list T2: Cue with the current context Work, hungry, 96th St There is a competition between S2 and S1: S2 was associated with 3/3 current contextual elements S1 was associated with 2/3 current contextual elements... so S2 wins the competition

33 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 Now we can begin to address delay effects in the Brown- Peterson paradigm S2 T2

34 Contextual Drift Switch at 96th St itchy Work hungry S1 9:40 9:41 9:42 9:43 This diagram illustrates the short-delay condition T1 S2 T2 S2 is recalled well because it matches the current context better than S1

35 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 S2 9:40 9:41 9:42 9:43 In this diagram, T2 is a long-delay trial If you wait long enough, most of the active contextual elements will change In this case, both S1 and S2 are equally poor matches to the current context (1/3 features), so recall will be poor T2

36 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 Summary: Items are associated with contextual features at study We use the current context as a retrieval cue S2 T2 According to the theory presented here, there is no effect of the passage of time per se: traces do not passively decay

37 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 S2 T2 9:40 9:41 9:42 9:43 Still, we end up getting better recall for recently presented items because the passage of time is correlated with contextual match Events that occurred more recently will have more contextual features in common with the retrieval cue, so their memory traces will be more strongly activated by the cue

38 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 S2 In the short-delay condition, contextual match to T2 is better for S2 vs. S1, so you get good recall T2

39 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 S2 9:40 9:41 9:42 9:43 In the long-delay condition, contextual match to T2 is equivalently poor for S2 vs. S1, so you get poor recall Telephone poles analogy (Crowder, 1976) T2

40 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 The fact that delay does not affect recall on the first trial can be explained in terms of the idea that recall is a competitive process Because memory is a competitive process, what matters most is the relative contextual match between S1 and T, rather than the absolute level of match

41 Contextual Drift Switch at 96th St itchy Work hungry S1 T1 9:40 9:41 9:42 9:43 Increasing delay reduces the absolute amount of study-test contextual match (in this case, from 3 features to 2) However, relative to other traces in your brain (relating to other things, besides Brown-Peterson) S1 is still by and far away the best-matching trace, so recall should be good

42 Summary - You can get effects of temporal delay without a decay process, if you assume that contextual drift occurs with time - However, we also still know that post-encoding processes that enhance memory stabilization DO occur (consolidation)..

43 + Overall Summary 28 n Reasons for forgetting over time: n n Interference: Learning about one thing can mess up your memory for other, similar things n Cue Overload Principle: The more competitors attached to a cue, the worse recall for any one item becomes n Proactive Interference: Previously encoded memories can disrupt retrieval of recently learned ones n Retroactive Interference: New learning can impair retrieval of older memories, either by structurally damaging these older memories or by competing with these older memories at retrieval Poor match between retrieval cues (at test) at processing at study n Contextual fluctuation: Subjects mental context is constantly changing; if your mental context now is very different from the way it was at study, this will hurt recall.

44 + Summary encoding 28 n Factors that lead to good memory n n Elaborative/distinctive encoding, which reduces interference and ensures that memory traces are available for recall at test Good match between retrieval cues and stored traces n The mnemonic techniques that we discussed use a structured retrieval framework to ensure that subjects can find (access) the memory traces that they formed at study...

45 5 Context and Retrieval 45

46 Memory Search Free recall delay etc Cartoon by Sean Polyn

47 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43

48 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43 shark computer bat test

49 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43 shark computer bat test Initiate recall by cuing with the current context: hungry, 96th St Given this cue, you end up recalling bat You also recall other contextual elements associated with bat: exams

50 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43 shark computer bat test

51 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43 shark computer bat test Step 2: Take retrieved contextual elements and incorporate them into your retrieval cue: hungry, exams, 96th St With exams in your retrieval cue, you can now recall computer, plus a new contextual element itchy

52 Reinstating Context 96th St itchy exams hungry 9:40 9:41 9:42 9:43 shark computer bat test Step 3: Incorporate itchy in your retrieval cue Now you can recall shark Using retrieved context as a retrieval cue allows you to bootstrap your way backwards in time...

53 Contextual Reinstatement What is the empirical evidence for contextual reinstatement?

54 Conditional Response Probability 1st 2nd 3rd 4th 5th 6th 7th notebook skull leaf watch shark bat computer

55 Conditional Response Probability 1st 2nd 3rd 4th 5th 6th 7th notebook skull leaf watch shark bat computer

56 Conditional Response Probability 1st 2nd 3rd 4th 5th 6th 7th notebook skull leaf watch shark bat computer Given that you just recalled the 4th item on the study list, what are the odds that the next item you recall will be the 3rd, or 5th, or 2nd, or 6th...?

57 Conditional Response Probability lag: st 2nd 3rd 4th 5th 6th 7th notebook skull leaf watch shark bat computer Conditional Response Probability lag (Graph from Kahana, 1996)

58 Conditional Response Probability lag: st 2nd 3rd 4th 5th 6th 7th notebook skull leaf watch shark bat computer Conditional Response Probability lag Key finding: Successive recalls come from nearby serial positions, in both the forward and backward directions (Graph from Kahana, 1996)

59 Conditional Response Probability Can this pattern be explained in terms of conjoint rehearsal?

60 Conditional Response Probability Can this pattern be explained in terms of conjoint rehearsal? No, it is found even when a long, arithmetic-filled delay is placed between items at study So what is going on?

61 Reinstating Context 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer

62 Reinstating Context 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer

63 Reinstating Context 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer

64 Reinstating Context 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer

65 Reinstating Context 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer

66 Recency Effect 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test Recency effect: Better recall of items from the end of the list Hypothesis: This occurs because contextual match is better for items from the end of the list

67 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test A slightly different task: judgments of recency

68 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test Which was presented more recently: shark or skull? Use the words to cue for contextual info

69 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test shark retrieves hungry, weather, exams, 96th St skull retrieves exams, itchy, 96th St compare retrieved context to current context: weather, 96th St

70 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test shark: hungry, weather, exams, 96th St skull: exams, itchy, 96th St current: weather, 96th St Contextual match is higher for shark, so shark is probably more recent...

71 Rats can be trained to do recency judgments Recording brain data from rats can give us a window into the role that contextual drift plays in recency judgments...

72 Show rats a series of odors Train rats to perform recency judgments Record multi-unit activity from region CA1 of the hippocampus

73 Treat this neural recording data as the animal s mental context Measure how much the neural activity pattern drifts during the encoding phase Use this to predict accuracy Intuitively: The more that mental context changes between item presentations, the more temporally discriminable the items will be

74 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test Here, context changes sharply between shark and computer In this situation, the shark vs. computer recency judgment is easy

75 Recency Judgments 96th St itchy exams hungry weather 9:40 9:41 9:42 9:43 notebook skull leaf watch shark bat computer test Here, context does not change at all between shark and computer In this situation, the shark vs. computer recency judgment is hard

76

77 Distance related to memory The experiment works! Increased neural drift between items predicts increased accuracy on the recency judgment Measuring neural drift gives us a window into how much the animal s mental context is drifting Which allows us to predict memory performance!77

78 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

79 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

80 from Squire & Kandel (1999)

81 from DeArmond, Fusco, & Dewey (1989)

82 Medial Temporal-lobe Memory System Hipp! ErC! PrC! How does this system contribute to episodic memory formation?!

83

84

85

86

87

88 The real deal...

89 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

90 + How does the Hippocampus Work? The job of the hippocampus is to rapidly memorize patterns of cortical activity (in entorhinal cortex?) so they can be recalled, given partial cues.. Key property: hippocampus tries to assign a distinct set of neurons to each memory This pattern separation property allows the hipp to learn rapidly without suffering catastrophic interference Goal for the first part of this lecture: Overview of how the hippocampus stores & retrieves memories, and how it accomplishes pattern separation

91 + Hippocampal binding STUDY HIPPO EC (input) EC (output)

92 + Hippo binding Activation Spreads from EC to Hipp (both directly and indirectly) STUDY HIPPO EC (input) EC (output)

93 + Hippo binding STUDY Recurrent connections in HIPP allow for Hebbian mechanisms of synaptic strengthening to occur HIPPO EC (input) EC (output)

94 + Hippo binding STUDY HIPPO EC (input) EC (output)

95 + Hippo binding STUDY Hippocampal episodic representation can then be read out into an output layer in EC HIPPO EC (input) EC (output)

96 + Pattern Completion TEST HIPPO EC (input) EC (output)

97 + Pattern Completion TEST HIPPO EC (input) EC (output)

98 + Pattern Completion TEST HIPPO EC (input) EC (output)

99 + Pattern Completion TEST HIPPO EC (input) EC (output)

100 + Pattern Separation STUDY HIPPO EC (input) EC (output)

101 + Explaining Pattern Separation How does the hippocampus manage to assign distinct representations to similar inputs? Answer: Sparse activity Only a small number of neurons are allowed to be active in hippocampus (relative to cortex) Sparse activity is enforced by inhibitory interneurons Inhibitory competition is especially fierce in the hippocampus When activity is sparse, small changes in the input pattern can have large effects on the hippocampal representation

102 + Pattern Separation and Sparse Activity Here, hippocampal units are connected to 5 inputs; only one hippocampal unit is allowed to be active at a time In the left-hand picture, the middle unit wins because all 5 of its inputs are active (whereas only 4 of its neighbor s inputs are active) Changing one input unit causes a different hippo unit to win!

103 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

104 + Large MTL damage in humans..like!104 H.M. n Everything we talked about before? n PLUS, spatial memory, implicit memory for associations, novel relational imagery

105 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

106 + Focal Hipp Lesions All of the amnesia data we have discussed thus far relates to big lesions, encompassing several MTL brain structures (e.g., perirhinal cortex) in addition to the hippocampus Recently, researchers have started to look at the effects of smaller lesions, limited to the hippocampus

107 + Outline The goal of this next section is to gain a better understanding of the respective contributions of the hippocampus vs. surrounding medial temporal lobe structures (perirhinal, parahippocampal, entorhinal cortex) Focus on 3 areas Fact learning Item recognition Association formation

108 + Vargha-Khadem et al. (1997) Vargha-Khadem et al. (1997)- looked at three patients (including Patient Jon ) who suffered hippocampal damage in childhood due to oxygen deprivation; surrounding cortical regions were spared Recall of once-presented stimuli is terrible study: window-reason; test: window-? However, these patients do OK in school...?

109 + Episodic/semantic Vargha-Khadem et al. (1997) talk about these findings in terms of a dissociation between episodic memory and semantic memory Episodic memory = recall & recognition of events Semantic memory = recall & recognition of facts Based on OK school performance, they argue that semantic learning ability may be intact However, this conclusion is premature.. Well-controlled studies of semantic learning in these patients have revealed that semantic learning is impaired also (e.g., Gardiner, Brandt, Baddeley, Vargha-Khadem, & Mishkin, 2008, Neuropsychologia)

110 + Episodic/semantic Episodic and semantic memory are closely intertwined Contrary to claims made in the Vargha-Khadem article, it is impossible to knock out episodic memory without there being some effect on semantic memory All fact memories start out as (hippocampally-dependent) episodic memories If a person is exposed to the fact many times (either because of many real world exposures, or because of hippocampal replay), the fact eventually sinks in to cortex and becomes a semantic memory.

111 + Explaining the Vargha-Khadem Results Overall: The patients good performance in school validates the idea that cortex is an enormously powerful learning system when given repeated exposure to study materials... Studies of patients with big lesions showed that, if you have some sparing of cortex, you can do some memorization of new facts The Vargha-Khadem study shows that, if you have a lot of sparing of cortex, you can do a lot of memorization of new facts

112 + Item Recognition In the patients studied by, Vargha-Khadem et al. (1997), item recognition is OK, even after 1 study exposure study: banana moose telephone test: eardrum? banana? Animal studies also support the idea that focal hippocampal damage does not disrupt item recognition In animals, recognition memory is typically tested using a delayed nonmatching to sample (DNMS) procedure

113 +

114 +

115 + Item Recognition If the hippocampus is not the key structure for item recognition memory, which structure(s) is most responsible for item recognition? Perirhinal cortex! Big lesions that encompass perirhinal cortex lead to poor recognition memory; subjects with spared perirhinal cortex show better item recognition memory

116 +

117 + Item Recognition Puzzle: How do we explain the finding of good recognition memory (after a single study exposure) in people and animals with focal hippocampal damage? The cortex learns incrementally Good recognition in these subjects after one study trial seems to contradict this incremental learning principle

118 + Perirhinal lower-level neocortex Even though cortex learns incrementally, repeated presentation of a stimulus leads to predictable changes in the overall structure of cortical representations...

119 + Cortical Recognition Repeating stimuli makes their representations sharper => novel stimuli weakly activate a large number of units => repeated stimuli strongly activate a small number of units novel --> <-- repeated There is less overall activity in the upper (perirhinal) layer for repeated vs. novel stimuli

120 + Recognition in Cortex The model s prediction of less activity (overall) in perirhinal cortex for repeated vs. novel stimuli has been validated by neuroimaging studies and single-cell-recording studies (e.g., Xiang & Brown, 1998) This finding of less activity for repeated stimuli implies that we can make recognition judgments by reading out the overall amount of activity in perirhinal cortex if activity is low, say studied if activity is high, say nonstudied

121 + Recognition in Cortex The idea that subjects base their recognition judgments on perirhinal activity is supported by a fmri study conducted by Gonsalves et al. (2005, Neuron)

122 + Relational Memory If the hippocampus isn t directly responsible for item recognition memory, what is the hippocampus responsible for? Eichenbaum and Cohen have argued that the hippocampus is crucial for relational memory: Rapidly forming a network of links between stimuli that can be flexibly traversed (in mind) by subjects...(think Tolman!)

123 +

124 + Konkel et al. (2009) Relational Memory Study The goal of this study was to simultaneously test item memory and different forms of relational memory in patients with focal hippocampal damage, patients with larger MTL lesions, and controls Predictions: Patients with focal hipp damage will show impaired relational memory and relatively spared item memory Patients with larger MTL lesions will show impaired relational AND item memory

125 +

126 +

127 + Subsequent Memory for Item and Relational Information FMRI studies have used the subsequent memory paradigm to explore how different brain regions contribute to encoding of item and relational information Scan people during study Later memory test Sort brain data from the study phase based on memory performance on the test phase (i.e. subsequently remembered, or subsequently forgotten) Compare the subsequently remembered brain scans to the subsequently forgotten brain scans

128 + Subsequent Memory for Item and Relational Information Davachi et al, (2003) used this paradigm to ask two questions: Activity in which brain region(s) predict subsequent item memory? Activity in which brain region(s) predict subsequent relational memory?

129 +

130 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of large MTL lesions - effects of more focal hippocampal lesions - roles of particular hippocampal substructures Perirhinal Ctx: item recognition, sharper signal with repetition Hippocampus: associative memory via pattern separation and completion

131 + Unitization and Associative Memory Just an aside: there are some circumstances where it appears that cortex (in particular, perirhinal cortex) can rapidly (i.e., based on one study trial) form new associations on its own Specifically: If subjects can unitize the two things being associated, then perirhinal cortex can form the association on its own

132 + The Medial Temporal Lobe and Memory - review of relevant anatomy (how hippo connects to cortex) - how the hippocampus works - effects of BIG lesions to the hippocampal region - effects of more focal hippocampal lesions - roles of particular hippocampal substructures

133 +!133

134 + Norman and O Reilly (2003)!134

135 + Pattern Completion TEST HIPPO EC (input) EC (output)

136 + Pattern Separation STUDY HIPPO EC (input) EC (output)

137 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

138 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

139 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

140 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

141 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

142 + DG = Pattern Separation Turbocharger DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

143 + Evidence for DG Pattern Separation Recently, researchers have started to look for direct evidence that DG is involved in pattern separation

144 + Leutgeb et al. (2007, Science) Leutgeb et al. (2007) manipulated the shape of a rat cage and recorded hippocampal activity in the dentate gyrus and region CA3 Prediction: The idea that dentate gyrus is especially prone to pattern separation implies that small changes in the input will have a larger effect on dentate gyrus activity than CA3 activity

145 +

146 + Bakker et al. (2008, Science) Bakker et al. used high-resolution fmri to study pattern separation in different hippocampal subregions Logic of the study is based on repetition suppression effects in fmri data

147 + Bakker et al. (2008, Science) If the same stimulus is presented twice, the fmri response is smaller on the second presentation If very different stimuli are presented, the fmri response stays constant This implies that we can use the size of the decrease in the fmri response as an indicator of whether a particular brain area thinks that two stimuli are the same or different If the two stimuli are coded in a similar fashion, we should see a decrease in the fmri response If the two stimuli are coded very differently, the fmri response should stay relatively constant

148 + Bakker et al. (2008, Science) In the Bakker et al. study, subjects performed a continuous recognition memory task Sometimes they presented lures that were very similar (but not identical) to studied items

149 + Bakker et al. (2008, Science) Predictions: Most brain regions should assign very similar representations to studied items and similar lures For these brain regions, we should see repetition attenuation (decreased responding to similar lures) DG should assign relatively dissimilar representations to studied items and similar lures For these brain regions, we should see less repetition attenuation (less of a decrease in responding to similar lures)

150 +

151 +DG is good for pattern separation, but it s bad for recall DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

152 + DG is good for pattern separation, but it s bad for recall DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

153 + DG is good for pattern separation, but it s bad for recall DG low overlap CA3 very low overlap high overlap EC (input) EC (output)

154 + Possible Solution: Two Modes DG low overlap CA3 Encoding mode: DG strong, facilitates pattern separation very low overlap high overlap EC (input) EC (output)

155 +Possible Solution: Two Modes DG low overlap CA3 Encoding mode: DG strong, facilitates pattern separation very low overlap Retrieval mode: DG weak, facilitates recall high overlap EC (input) EC (output)

156 +Possible Solution: Two Modes DG low overlap CA3 Encoding mode: DG strong, facilitates pattern separation very low overlap Retrieval mode: DG weak, facilitates recall high overlap EC (input) EC (output)

157 + Evidence for Mode-Setting There is evidence that acetylcholine might adjust the strengths of hippocampal pathways so as to facilitate encoding (as opposed to retrieval) For example, ACh blocks transmission along CA3 recurrents while still allowing learning in these synapses: Good for encoding Low ACh facilitates transmission along CA3 recurrents: Good for retrieval

158 + Evidence for Mode-Setting Hasselmo & Wyble (1997) set forth a detailed story about how novel stimuli could trigger ACh and flip the hippocampus into encoding mode Note that the time course of ACh release is relatively slow...

159 + Evidence for mnemonic modes?!159 (Duncan, Sadanand and Davachi, 2012 Science)

160 +!160

161 + Barense et al 2012!161

162 The take-home Computations supported by the hippocampus and cortical structures appear fundamentally distinct Recollection and familiarity are likely independent processes that support memory function Interplay between empirical findings in animal and human research with modeling has played a big role in memory research Has led to a focus on the computations performed in the brain - not so much on identifying a memory system or a perceptual system 162

163 +!163

164 + Neurogenesis in the Dentate Gyrus The dentate gyrus site of neurogenesis Hypothesis: New neurons help to randomize the dentate gyrus response, thereby ensuring that similar stimuli activate different sets of dentate gyrus neurons The idea that neurogenesis in the dentate gyrus helps pattern separation implies that blocking neurogenesis will hurt pattern separation

165 + Clelland Et Al. (2009, Science) Clelland et al. (2009) set out to explore how disrupting neurogenesis in the dentate gyrus affects learning They used a task where animals visited a place, and then they had to discriminate between the previously visited place and another place In some conditions, the animals had to discriminate between nearby places, and in other conditions the animals had to discriminate between far away places

166 + Clelland Et Al. (2009, Science) Key idea: Hippocampal pattern separation is important for discriminating between nearby places but it is not needed to discriminate between far away places Prediction: IF disrupting neurogenesis impairs pattern separation AND pattern separation is important for discriminating between nearby places THEN disrupting neurogenesis should impair discrimination between nearby places but it should not disrupt discrimination between far away places

167 +

168 + Summary!168

169 Key Principals for the Semester Learning and memory are closely related and intertwined states of information processing Major insights about learning and memory have come from studies of the brain The concept of multiple memory systems unifies the study of learning and memory The underlying bases of learning and memory are the same in humans and animals Our theoretical approaches to studying learning are always closely tied to technological advances that are unfolding in general society (e.g., today - machine learning) 169