Christopher M. Glaze

Temporal structure in zebra finch song : Implications for motor coding

Adult zebra finch songs consist of stereotyped sequences of syllables. Although some behavioral and physiological data suggest that songs are structured hierarchically, there is also evidence that they are driven by nonhierarchical, clock-like bursting in the premotor nucleus HVC (used as a proper name). In this study, we developed a semiautomated template-matching algorithm to identify repeated sequences of syllables and a modified dynamic time-warping algorithm to make fine-grained measurements of the temporal structure of song. We find that changes in song length are expressed across the song as a whole rather than resulting from an accumulation of independent variance during singing. Song length changes systematically over the course of a day and is related to the general level of bird activity as well as the presence of a female. The data also show patterns of variability that suggest distinct mechanisms underlying syllable and gap lengths: as tempo varies, syllables stretch and compress proportionally less than gaps, whereas syllable–syllable and gap–gap correlations are significantly stronger than syllable–gap correlations. There is also increased temporal variability at motif boundaries and especially strong positive correlations between the same syllables sung in different motifs. Finally, we find evidence that syllable onsets may have a special role in aligning syllables with global song structure. Generally, the timing data support a hierarchical view in which song is composed of smaller syllable-based units and provide a rich set of constraints for interpreting the results of physiological recordings.

Behavioral Measurements of a Temporally Precise Motor Code for Birdsong

There are conflicting data on the timescale for the representation of adult zebra finch song. Acoustic structure and perturbation studies suggest that song is divided into discrete vocal elements, or syllables, lasting 50–200 ms. However, recordings in premotor telencephalic nucleus HVC (used as proper name) and RA (robust nucleus of arcopallium) suggest that song is represented by sparse, fine-grained bursting on the 5–10 ms timescale. We previously found patterns of timing variability that distinguish individual syllables and repeat across multiple 500- to 1000-ms-long motifs (Glaze and Troyer, 2006). Here, we extend our methods to analyze whether this is attributable to a syllable-based code or representations on a finer timescale. We find evidence for the latter. First, identity-dependent timing is dominated by independent variability in notes, finer song segments that compose a syllable; for example, the length of a note is no more correlated with other notes in the same syllable than it is with notes in other syllables. For a subset of notes, clear modulation in spectral structure allowed for accurate timing measurements on the 5–10 ms timescale. Temporal independence holds at this scale as well: the length of an individual 5–10 ms song slice is correlated with the same slice repeated 500–1000 ms later, yet is independent of neighboring slices. We propose that such fine-grained, persistent changes in song tempo result from an interaction between slow modulatory factors and precisely timed, sparse bursting in HVC and RA.

Development of temporal structure in zebra finch song

Zebra finch song has provided an excellent case study in the neural basis of sequence learning, with a high degree of temporal precision and tight links with precisely timed bursting in forebrain neurons. To examine the development of song timing, we measured the following four aspects of song temporal structure at four age ranges between 65 and 375 days posthatch: the mean durations of song syllables and the silent gaps between them, timing variability linked to song tempo, timing variability expressed independently across syllables and gaps, and transition probabilities between consecutive syllable pairs. We found substantial increases in song tempo between 65 and 85 days posthatch, due almost entirely to a shortening of gaps. We also found a decrease in tempo variability, also specific to gaps. Both the magnitude of the increase in tempo and the decrease in tempo variability were correlated on gap-by-gap basis with increases in the reliability of corresponding syllable transitions. Syllables had no systematic increase in tempo or decrease in tempo variability. In contrast to tempo parameters, both syllables and gaps showed an early sharp reduction in independent variability followed by continued reductions over the first year. The data suggest that links between syllable-based representations are strengthened during the later parts of the traditional period of song learning and that song rhythm continues to become more regular throughout the first year of life. Similar learning patterns have been identified in human sequence learning, suggesting a potentially rich area of comparative research.

Normative evidence accumulation in unpredictable environments

In our dynamic world, decisions about noisy stimuli can require temporal accumulation of evidence to identify steady signals, differentiation to detect unpredictable changes in those signals, or both. Normative models can account for learning in these environments but have not yet been applied to faster decision processes. We present a novel, normative formulation of adaptive learning models that forms decisions by acting as a leaky accumulator with non-absorbing bounds. These dynamics, derived for both discrete and continuous cases, depend on the expected rate of change of the statistics of the evidence and balance signal identification and change detection. We found that, for two different tasks, human subjects learned these expectations, albeit imperfectly, then used them to make decisions in accordance with the normative model. The results represent a unified, empirically supported account of decision-making in unpredictable environments that provides new insights into the expectation-driven dynamics of the underlying neural signals. DOI: http://dx.doi.org/10.7554/eLife.08825.001

A Generative Model for Measuring Latent Timing Structure in Motor Sequences

Motor variability often reflects a mixture of different neural and peripheral sources operating over a range of timescales. We present a statistical model of sequence timing that can be used to measure three distinct components of timing variability: global tempo changes that are spread across the sequence, such as might stem from neuromodulatory sources with widespread influence; fast, uncorrelated timing noise, stemming from noisy components within the neural system; and timing jitter that does not alter the timing of subsequent elements, such as might be caused by variation in the motor periphery or by measurement error. In addition to quantifying the variability contributed by each of these latent factors in the data, the approach assigns maximum likelihood estimates of each factor on a trial-to-trial basis. We applied the model to adult zebra finch song, a temporally complex behavior with rich structure on multiple timescales. We find that individual song vocalizations (syllables) contain roughly equal amounts of variability in each of the three components while overall song length is dominated by global tempo changes. Across our sample of syllables, both global and independent variability scale with average length while timing jitter does not, a pattern consistent with the Wing and Kristofferson (1973) model of sequence timing. We also find significant day-to-day drift in all three timing sources, but a circadian pattern in tempo only. In tests using artificially generated data, the model successfully separates out the different components with small error. The approach provides a general framework for extracting distinct sources of timing variability within action sequences, and can be applied to neural and behavioral data from a wide array of systems.

Pulling an all-nighter

Research indicates that sleep influences learning, but little is known about the mechanisms involved. A recent article suggests that sleep modifies the firing patterns of sensorimotor neurons before there is improvement in performance.

A bias–variance trade-off governs individual differences in on-line learning in an unpredictable environment

Decisions often benefit from learned expectations about the sequential structure of the evidence. Here we show that individual differences in this learning process can reflect different implicit assumptions about sequence complexity, leading to performance trade-offs. For a task requiring decisions about dynamic evidence streams, human subjects with more flexible, history-dependent choices (low bias) had greater trial-to-trial choice variability (high variance). In contrast, subjects with more history-independent choices (high bias) were more predictable (low variance). We accounted for these behaviours using models in which assumed complexity was encoded by the size of the hypothesis space over the latent rate of change of the source of evidence. The most parsimonious model used an efficient sampling algorithm in which the range of sampled hypotheses represented an information bottleneck that gave rise to a bias–variance trade-off. This trade-off, which is well known in machine learning, may thus also have broad applicability to human decision-making.Glaze et al. show that individual variability in learning from noisy evidence involves a bias–variance trade-off that is best explained by a model using a sampling algorithm that approximates optimal inference.

Temporal variability in a synfire chain model of birdsong

Introduction Zebra finch songs are highly stereotyped, making them especially well suited for in depth analysis of the neural mechanisms underlying sequential behaviors. The acoustic structure of song is arranged into vocal units, known as syllables, which are ordered into highly invariant sequences called motifs. However, recordings in highlevel premotor nuclei during singing reveal spike bursts in individual neurons with two notable timing properties: first, burst times have millisecond precision, and second, the times have no obvious relationship to the timing of syllables or intersyllable gaps [1]. This has led to the hypothesis that the representation for song is clock-like, and is generated as neural activity propagates down a chain-like network of neurons known as a synfire chain [2,3].

Regulation of vocal precision by local noradrenergic modulation of a motor nucleus

Recent theories of norepinephrine (NE) function suggest that NE modulates the transition between stereotyped, goal-directed behavior and more variable exploratory behaviors that facilitate learning and adaptation. We provide evidence for context dependent switching by NE that is analogous to this explore/exploit strategy in the vocal system of the zebra finch (Taeniopygia guttata). Stimulation of the locus coeruleus, the major source of norepinephrine in the brain, decreases song trial-to-trial variability transforming the variable, exploratory “undirected” song into song that resembles the more stereotyped, exploitative “directed” song males sing to females. This behavioral switch is mediated by NE acting directly on a cortical motor nucleus that integrates inputs from a premotor cortical nucleus and a basal ganglia circuit necessary for vocal motor learning. These findings suggest that norepinephrine can act directly on the motor system to influence the transition between exploratory and exploitative behavioral strategies.