Mexico

Architecture search optimizes the structure of a neural network for some task instead of relying on manual authoring. However, it is slow, as each potential architecture is typically trained from scratch. In this paper we present an approach called Conceptual Expansion Neural Architecture Search (CENAS) that combines a sample-efficient, computational creativityinspired transfer learning approach with neural architecture search. This approach finds models faster than naive architecture search via transferring existing weights to approximate the parameters of the new model. It outperforms standard transfer learning by allowing for the addition of features instead of only modifying existing features. We demonstrate that our approach outperforms standard neural architecture search and transfer learning methods in terms of efficiency, performance, and parameter counts on a variety of transfer learning tasks.

ICCC&#39;21

Conceptual Expansion Neural Architecture Search (CENAS)

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible. The experience for on-site participants would closely approximate a normal pre-COVID *ACL conference, with 5-6 thematically organized parallel sessions and live Q/A and interactive discussion immediately after the talks. Presentations by virtual participants will be equitably interleaved with those of on-site participants, projected on the auditorium screens as if on-site, and also followed immediately by live Q/A and interactive discussion at a time during reasonable waking hours for the virtual presenter. For all participants, on-site and virtual, who are unable to attend a session due to either time-zone issues or because they are participating in another session live, talk recordings and slides will be available online at a minimum after the live presentation (and in many cases before as well), and questions may be submitted in advance on session-specific discussion boards and answered live in session with the usual visual aids if desired.

<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/ceby/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

Please Note: The EMNLP registration system is not currently connected to the underline site as we are still in the process of building out EMNLP 2021. You will receive access instructions from underline the week of November 1st. 

Access is given only to EMNLP upon registration, if you have not registered please do so [here](https://2021.emnlp.org/registration).

Registered attendees will receive access the week of November 1st.

EMNLP 2021

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible.

poster

### **Welcome to the Twelfth International Conference on Computational Creativity, ICCC’21!**

<iframe width=1280 height=720 style="border:0;" scrolling="no" src="https://screencast-o-matic.com/player/crQbitV66dw?width=1280&height=720&ff=1&title=0" allowfullscreen="true"></iframe>

Computational creativity is an interdisciplinary endeavor that is located at the intersection of the fields of artificial intelligence, cognitive psychology, philosophy, social sciences and the arts. The goal of computational creativity is to model, simulate or replicate creativity using a computer, to achieve one of several ends:

* to construct a program or computer capable of showing some level of creativity
* to better understand human creativity and to formulate an algorithmic perspective on creative behavior in humans
* to design programs that can enhance human creativity without necessarily being creative themselves

<br>The ICCC conference series organized by the [**Association for Computational Creativity**](https://computationalcreativity.net/) since 2010 is the only scientific conference that focuses on computational creativity alone and also covers all aspects of it.



To access the Twelfth International Conference on Computational Creativity, ICCC’21, you need to register for the conference.  Instructions on how to access the event will be provided directly to attendees and included on the Twelfth International Conference on Computational Creativity, ICCC’21 conference website. 

[Register now](https://computationalcreativity.net/iccc21/registration/)

Register now

ICCC'21

Welcome to the Twelfth International Conference on Computational Creativity, ICCC’21!

Large pre-trained transformer-based language models have revolutionized the field of natural language processing in recent years. While BERT-like models perform exceptionally well for analytical tasks such as classification and regression, their text generation capabilities are usually limited to predicting tokens within a given context. In this paper, we introduce Galmet, a model that generates text by using genetic algorithms with BERT-like language models for evolving text. We use Galmet with the RoBERTa language model to automatically evolve real headlines into more satirical headlines. This is achieved by adapting the masked language head to the headlines domain for the mutation operator and finetuning a regression head to distinguish headlines from satire for the fitness function. We evaluated our system by comparing generated satirical headlines against human-edited headlines and just the fine-tuned masked language head. We found that while humans generally outperform the model, generations by Galmet are also often preferred over human-edited headlines. However, we also found that only using the fine-tuned masked language model gives slightly preferred satire due to generating more readable sentences. Galmet is thus a first step towards a new way of creating text generators using masked language models by transforming text guided by scores from another language model.

Survival of the Wittiest: Evolving Satire with Language Models

Recent interest in story generators capable of combining more than on plot line into an elaborate story have been handicapped by the lack of either theoretical material or quantitative metrics to ascertain the quality of outputs of such attempts. The present short paper postulates a set of metrics designed to capture some of the insights elaborated during a formative evaluation of an existing attempt at plot weaving.

Assessing MultiPlot Stories: from Formative Analysis to Computational Metrics

This paper concerns the evaluation of a workspace architecture for generating natural language descriptions, including methods for evaluating both its output and its own self-evaluation. Herein are details of preliminary results from evaluation of an early iteration of the architecture operating in the domain of weather. The domain is not typically seen as creative, but provides a simple testbed for the architecture and evaluation methodology. The program does not yet match humans in terms of fluency of language, factual correctness, and how completely the input is described, but human judges did find the program's output easier to read than human generated texts. Planned improvements to the program also described in the paper will incorporate self-monitoring and better self-evaluation with the aim of producing descriptions that are more fluently written and more accurate.

Evaluating Natural Language Descriptions Generated in a Workspace-Based Architecture

RoboCHAIR is a system for generating scientific questions and is implemented as a web interface. We focus on generating questions which can trigger new scientific
ideas or bring attention to the elements that need clarifications. This paper extends the initial version of the RoboCHAIR’s question generation module by keeping only a selection of best evaluated templates from the initial pattern-based approach and proposes a novel method based on triplet graphs enriched with word embeddings
to identify parts of text which require clarifications by the author(s). In this paper the two methods are compared showing that the pattern-based method achieves higher scores.

Scientific question generation: pattern-based and graph-based RoboCHAIR methods

Social media is now the megaphone of choice for many digital-age provocateurs. Social networks offer a wealth of examples of egregious misbehaviour by humans, but perhaps more worryingly, they also serve to magnify and weaponize the automated outputs of generative systems. So what responsibilities do the creators of these systems bear for the offenses caused by their creations, and what can they do to prevent, or mitigate, the worst excesses? For those who “make things that make things,” the issue is further complicated by the potential of our systems to accidentally tread on the feelings of others. When does a norm become a troubling stereotype, and how should a system built on social norms avoid crossing the line? We argue that because this line is so easy to cross, our systems should aim to engage with and mitigate offense rather than try vainly to prevent it completely.


Walking the Line in Computational Creativity: Paradox and Pitfall in the Mitigation of Automated Offense

Although creativity is an ability that can be developed if trained, most of the systems developed for creativity stimulation focused on individual interventions. This work presents the design, development, and evaluation of CUBUS, a digital tool developed to inspire social creativity in children through storytelling and emotionally expressive characters. Groups of children collaborate with the autonomous virtual characters of the digital tool to create a story together. Our main goal was to conduct an experimental study (n=20) to investigate if the interaction between children and the autonomous emotionally expressive virtual characters present in CUBUS can stimulate social creativity during a storytelling activity. We measured the impact of the digital tool across several creativity domains: pre-and post-testing (questionnaire assessment), the creative process of children (behavior analysis), and the story created (creative outcome). Results showed that although children generated fewer storytelling ideas using CUBUS, these ideas were deemed more original.

Inspiring Social Creativity in Children with a Digital Tool

What if creative machines could show the story told by a live orchestra? In this paper, we propose a vision for a co-creative orchestra performance partner that enlivens this classical art form and has the potential to increase audience engagement and reach broader audiences. We present MIST, an early version of our vision, which creates real-time visualizations as live music is performed, and showcase it in a violin performance.

MIST: ``You Play, I'll Draw'

The ability to produce surprising outputs is a cornerstone of creative behavior. In this paper we propose deep learning architectures that are trained in an adversarial setting to learn patterns of expectations and surprise from data. We introduce benchmark datasets of geometrical shapes that represent well defined patterns of violations of expectations, and use them to verify empirically that the separation between learning of expectations and learning of surprise is essential for achieving good generalization performance.

Adversarial Learning of Expectation and Surprise: Experiments with Geometric Shapes

The study of Ukiyo-e, an important genre of pre-modern Japanese art, focuses on the object and style like other artwork researches. Such study has benefited from the renewed interest by the machine learning community in culturally important topics, leading to interdisciplinary works including collections of images, quantitative approaches, and machine learning-based creativities. They, however, have several drawbacks, and it remains challenging to integrate these works into a comprehensive view. To bridge this gap, we propose a holistic approach We first present a large-scale Ukiyo-e dataset with coherent semantic labels and geometric annotations, then show its value in a quantitative study of Ukiyo-e paintings' object using these labels and annotations. We further demonstrate the machine learning methods could help style study through soft color decomposition of Ukiyo-e, and finally provides joint insights into object and style by composing sketches and colors using colorization

Ukiyo-e Analysis and Creativity with Attribute and Geometry Annotation

Metaphor is a powerful tool in the performer’s tool box, not least because it can operate at several levels at once. As our linguistic metaphors deliver rhetorical flourishes, conceptual metaphors change the way we see the world, while our metaphorical body language allows us to take postures and stances that are both figurative and literal. Each kind of metaphor is another expressive choice that creative performers can make to convey their meanings. This choice sits at the heart of what it means to be a creative agent, for agents who lack choice and do exactly as they are told cannot exercise genuine intent for their actions. We shall explore how interpretation is wrapped up with choice to appreciate and to achieve emotional creativity in a system for generating and enacting automated tales. We explain here how three levels of description – linguistic, conceptual and physical – are integrated in a framework that motivates the use of metaphor or irony as a creative choice, by robot performers aiming to go beyond a predetermined script.

Conceptual Expansion Neural Architecture Search (CENAS)

Downloads

Next from ICCC'21

Survival of the Wittiest: Evolving Satire with Language Models

Similar lecture

Language, Brains & Interpretability

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES