We present a novel retrofitting method to induce emotion aspects into pre-trained language models (PLMs) such as BERT and RoBERTa.Our method updates pre-trained network weights using contrastive learning so that the text fragments exhibiting similar emotions are encoded nearby in the representation space, and the fragments with different emotion content are pushed apart.While doing so, it also ensures that the linguistic knowledge already present in PLMs is not inadvertently perturbed.The language models retrofitted by our method, i.e., BERTEmo and RoBERTaEmo, produce emotion-aware text representations, as evaluated through different clustering and retrieval metrics.For the downstream tasks on sentiment analysis and sarcasm detection, they perform better than their pre-trained counterparts (about 1% improvement in F1-score) and other existing approaches.Additionally, a more significant boost in performance is observed for the retrofitted models over pretrained ones in few-shot learning setting.

Word embeddings learned using the distributional hypothesis (e.g., GloVe, Word2vec) are good at encoding various lexical-semantic relations. However, they do not capture the emotion aspects of words. We present a novel retrofitting method for updating the vectors of emotion bearing words like fun, offence, angry, etc. The retrofitted embeddings achieve better inter-cluster and intra-cluster distance for words having the same emotions, e.g., the joy cluster containing words like fun, happiness, etc., and the anger cluster with words like offence, rage, etc., as evaluated through different cluster quality metrics. For the downstream tasks on sentiment analysis and sarcasm detection, simple classification models, such as SVM and Attention Net, learned using our retrofitted embeddings perform better than their pre-trained counterparts (about 1.5 % improvement in F1-score) as well as other benchmarks. Furthermore, the difference in performance is more pronounced in the limited data setting.

Artificial intelligence and machine learning are being increasingly used in scientific domains such as computational fluid dynamics and chemistry. Particularly notable is a recently renewed interest in solving partial differential equations using machine learning models, especially deep neural networks, as partial differential equations arise in many scientific problems of interest. Within materials science literature, there has been a surge in publications on AI-enabled materials discovery, in the last five years. Despite this, the interaction between machine learning researchers and materials scientists (especially, scientists working on structural materials, their microstructures, textures and so on) has been very sparse. On the other hand, AI/ML techniques can clearly be integrated into materials design frameworks (e.g., MGI efforts) to support accelerated materials development, novel simulation methodologies and advanced data analytics. Hence there is an immediate need for exchange of ideas and collaborations between machine learning and materials science communities. We believe a workshop dedicated to this theme would be well- suited to foster such collaborations. The aim of this workshop is to bring together the computer science and materials science communities and foster deeper collaborations between the two to accelerate the adoption of AI/ML in materials science. We hope and envision this workshop to facilitate in building a community of researchers in this interdisciplinar area in the years ahead.

Devices like solar cells, batteries etc. often comprise of a host of material types including organic, inorganic and hybrid materials. The fabrication procedures for these devices involve screening or designing the right set of materials and then subjecting them to a sequence of operations under very specific conditions. The performance characteristics of a device critically depend on the materials used in its fabrication, the specific operations carried out, their operating conditions and the specific sequence in which they are carried out. The space of potential materials, operations and operating conditions is vast, and selecting the right combination thereof to achieve the desired characteristics is a knowledge intensive activity. A large amount of such device fabrication knowledge is available in the form of publications, patents, company reports and so on. In this paper, we present a system that systematically extracts this knowledge from materials science literature. The extracted knowledge is represented as knowledge graphs conforming to an ontology that can be queried to make informed decisions in device fabrication procedures. The system first identifies the set of relevant paragraphs that contain fabrication knowledge. It then employs state of the art entity and relation extraction models to identify instances of operations, methods, materials, etc. and relations between them. The system then applies an unsupervised algorithm to identify sequences of operations representing fabrication procedures. We applied our system on solar cell fabrication knowledge extraction and achieved good performance. We believe our results provide much needed impetus for further work in this area.

We present a novel retrofitting model that can leverage relational knowledge available in a knowledge resource to improve word embeddings. The knowledge is captured in terms of relation inequality constraints that compare similarity of related and unrelated entities in the context of an anchor entity. These constraints are used as training data to learn a non-linear transformation function that maps original word vectors to a vector space respecting these constraints. The transformation function is learned in a similarity metric learning setting using Triplet network architecture. We applied our model to synonymy, antonymy and hypernymy relations in WordNet and observed large gains in performance over original distributional models as well as other retrofitting approaches on word similarity task and significant overall improvement on lexical entailment detection task.

Physics based simulations of manufacturing processes are used for prediction of material properties and defects in a number of industrial applications. However, a practising engineer often requires the solution to an "inverse problem" - prediction of inputs for the desired outcome. The inverse problem is usually solved by constrained optimisation. Extensive simulation during optimisation is avoided through response surfaces constructed from simulations. But the design space is often so large that even with response surfaces, optimisation might not be possible. Moreover, these problems are typically ill-posed, so discriminative models such as artificial neural networks do not work well. In this paper, we investigate the application of conditional linear Gaussian Bayesian networks to address the inverse problem with multi-pass wire drawing process as a case study. We propose an approach to systematically find all solutions and rank them according to their likelihood.

Integrated computational materials engineering (ICME) is a new approach to the design and development of materials, manufacturing processes and products.The approach proposes using a combination of modeling and simulation, datadriven reasoning and knowledge-guided decision making.Industrialization of this approach requires strong automation support.Modeling and simulation is a highly knowledge-intensive activity and integrated design requires knowledge cutting across several design domains.For the industrialization vision to succeed, it is essential to capture this knowledge and make it available in a usable form for people not so skilled in these areas.With this motivation, we are building a comprehensive computational platform to support this emerging design paradigm.We present an overview of the platform and discuss its knowledge engineering requirements.We present a knowledge modeling framework to address these requirements.

We discuss our experience of using model-based techniques to generate model-based code generators. The central idea behind model-driven development (MDD) is to use suitable models to specify various concerns and transform these models to a variety of text artifacts. A business product needs to deliver a given business functionality on a wide variety of implementation platforms and architectures thus necessitating multiple sets of code generators. However, there is a considerable commonality across these code generators. In absence of a suitable abstraction for capturing this commonality, there is little or no reuse across these code generators. We present an abstraction for organizing model-based code generators as a hierarchical composition of reusable building blocks. A building block is a localized specification of a concern in terms of a concern-specific meta model, model to model trans-formation, and model to text transformation. Model-based code generation is a 3-step walk over the composition tree wherein the first step transforms individual concern-specific models into a unified model, the second step transforms the unified model into individual concern-specific text artifacts, and the third step composes these text artifacts.

We discuss our experience of introducing MDA approach, supported by our MDA toolset, in an IT consultancy and software development services organization. Initially, our MDA toolset provided a set of modeling notations for specifying different layers of an enterprise application and a set of code generators that transform these models into platform-specific implementations. In spite of considerable amount of code being generated, it didn't translate into commensurate productivity gain over the entire development lifecycle of large projects. What was missing was support for collaborative development. Providing suitable abstractions and tool support for a collaboration-based development method improved the situation considerably. Enterprise product lines and large development projects benefited the most resulting in improved productivity, better code quality, platform independence and easier change management. However, the steep learning curve and inadequate debugging support for models were significant hurdles. Small to medium sized projects found this MDA approach too heavy and restrictive, and preferred a flexible and lightweight meta-data driven aspect-oriented approach. Our experience suggests that MDA will find greater acceptance if complimented with well-defined and flexible methodologies that support collaborative development.