Phil 11.28.17

7:00 – 8:00 Research, 8:30 – 4:30 ASRC MKT

  • Continuing Speaker–listener neural coupling underlies successful communication here. Done! Will promote to Phlog later.
  • Collective cognition in animal groups
    • Iain Couzin
    • The remarkable collective action of organisms such as swarming ants, schooling fish and flocking birds has long captivated the attention of artists, naturalists, philosophers and scientists. Despite a long history of scientific investigation, only now are we beginning to decipher the relationship between individuals and group-level properties. This interdisciplinary effort is beginning to reveal the underlying principles of collective decision-making in animal groups, demonstrating how social interactions, individual state, environmental modification and processes of informational amplification and decay can all play a part in tuning adaptive response. It is proposed that important commonalities exist with the understanding of neuronal processes and that much could be learned by considering collective animal behavior in the framework of cognitive science.
  • The Socio-Temporal Brain: Connecting People in Timethis looks like it might explicitly link human neural coupling and flocking.
    • Temporal and social processing are intricately linked. The temporal extent and organization of interactional behaviors both within and between individuals critically determine interaction success. Conversely, social signals and social context influence time perception by, for example, altering subjective duration and making an event seem ‘out of sync’. An ‘internal clock’ involving subcortically orchestrated cortical oscillations that represent temporal information, such as duration and rhythm, as well as insular projections linking temporal information with internal and external experiences is proposed as the core of these reciprocal interactions. The timing of social relative to non-social stimuli augments right insular activity and recruits right superior temporal cortex. Together, these reciprocal pathways may enable the exchange and respective modulation of temporal and social computations.
    • timing is not encapsulated but interacts closely with the social processes that emerge from interpersonal interactions [1–3]. As interactional behaviors play out in time, their temporal signatures carry important information. They guide attention, convey a message, and mold bonds between individuals. Moreover, interactional behaviors in turn give meaning to time and influence its perception and representation. A range of disorders that jointly compromise temporal and social processes attest to this relation (pp 760)
    • We review here the many ways in which timing intersects with social processing. We explore this intersection for the communicative behavior of an individual as well as for the behavioral coordination between communicating agents. We detail neuroimaging evidence on how temporal and social information are represented in the brain and identify points of structural and functional convergence (pp 760)
    • The temporal coordination of behavior in animals is referred to as chorusing. Chorusing describes sporadic behaviors such those of flying birds (Figure I), which may change direction and speed in a manner resembling a single superorganism [87]. In addition, it describes behaviors that occur repeatedly and with some amount of temporal regularity, as in the mating calls of male frogs. Although less frequent than sporadic chorusing, rhythmic chorusing is displayed by a range of taxa including insects, reptiles, birds, and mammals [88, 89]. (pp 761)
    • Research into the functionality of chorusing suggests species differences. For some, it seems to be a mere byproduct of competitive interactions. For others, it reduces the risk of predation [98]. By analogy with the selfish-herd principle, overlapping with others in time makes it harder for predators to single out individual prey. Last, there are species in which chorusing serves as a fitness display in the context of sexual selection [99] and as a means to foster social bonds [100]. Because of its pervasiveness and social context, some suggest chorusing to be the driving evolutionary force for a species’ timing sense [88]. (pp 761)
    • The degree of temporal coordination between interaction partners relates to interaction success. For example, it produces affective consequences [21]. This was revealed by a study in which pairs of strangers discussed four topics and completed an affective state questionnaire before and after each topic. Results provided evidence for emerging synchrony between discussion partners (Figure 1) and for its causal effect on ensuing positive affect. Related research showed that individuals more readily empathize with a synchronous as compared to a non-synchronous partner [22] and that synchronous dyads are more creative [23] and trusting [24] than nonsynchronous dyads. (pp 762)
  • Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypes
    • Word embeddings use vectors to represent words such that the geometry between vectors captures semantic relationship between the words. In this paper, we develop a framework to demonstrate how the temporal dynamics of the embedding can be leveraged to quantify changes in stereotypes and attitudes toward women and ethnic minorities in the 20th and 21st centuries in the United States. We integrate word embeddings trained on 100 years of text data with the U.S. Census to show that changes in the embedding track closely with demographic and occupation shifts over time. The embedding captures global social shifts – e.g., the women’s movement in the 1960s and Asian immigration into the U.S – and also illuminates how specific adjectives and occupations became more closely associated with certain populations over time. Our framework for temporal analysis of word embedding opens up a powerful new intersection between machine learning and quantitative social science.
    • bias
    • I think this is really close to the belief trajectories I’m trying to tease out. In the figures above, note that it is possible to extract both trajectories and normative terms. Plus, the paper has a really good writeup of methods in the appendices.
  • Continuing to write up use cases/RB proposal
  • Also still doing stuff for the HHS RFI
  • Relevant to the Research Browser:
    • Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples <- Agent generation is a thing!
      • We address the problem of extracting an automaton from a trained recurrent neural network (RNN). We present a novel algorithm that uses exact learning and abstract interpretation to perform efficient extraction of a minimal automaton describing the state dynamics of a given RNN. We use Angluin’s L* algorithm as a learner and the given RNN as an oracle, employing abstract interpretation of the RNN for answering equivalence queries. Our technique allows automaton-extraction from the RNN while avoiding state explosion, even when the state vectors are large and fine differentiation is required between RNN states. 
        We experiment with automata extraction from multi-layer GRU and LSTM based RNNs, with state-vector dimensions and underlying automata and alphabet sizes which are significantly larger than in previous automata-extraction work. In some cases, the underlying target language can be described with a succinct automata, yet the extracted automata is large and complex. These are cases in which the RNN failed to learn the intended generalization, and our extraction procedure highlights words which are misclassified by the seemingly “perfect” RNN.
    • Policy Shaping: Integrating Human Feedback with Reinforcement Learning
      • A long term goal of Interactive Reinforcement Learning is to incorporate nonexpert human feedback to solve complex tasks. Some state-of-the-art methods have approached this problem by mapping human information to rewards and values and iterating over them to compute better control policies. In this paper we argue for an alternate, more effective characterization of human feedback: Policy Shaping. We introduce Advise, a Bayesian approach that attempts to maximize the information gained from human feedback by utilizing it as direct policy labels. We compare Advise to state-of-the-art approaches and show that it can outperform them and is robust to infrequent and inconsistent human feedback.
    • AXIS: Generating Explanations at Scale with Learnersourcing and Machine Learning
      • While explanations may help people learn by providing information about why an answer is correct, many problems on online platforms lack high-quality explanations. This paper presents AXIS (Adaptive eXplanation Improvement System), a system for obtaining explanations. AXIS asks learners to generate, revise, and evaluate explanations as they solve a problem, and then uses machine learning to dynamically determine which explanation to present to a future learner, based on previous learners’ collective input. Results from a case study deployment and a randomized experiment demonstrate that AXIS elicits and identifies explanations that learners find helpful. Providing explanations from AXIS also objectively enhanced learning, when compared to the default practice where learners solved problems and received answers without explanations. The rated quality and learning benefit of AXIS explanations did not differ from explanations generated by an experienced instructor.