Soo-Young Lee

AI based design for OLED materials are being tried in a variety of ways. An exemplary system is being developed to predict optical characteristics through machine learning (ML) with existing data. Once the performance descriptor is well defined and the quantum chemical calculation method is established, AI‐reverse design is expected to be possible. However, not all OLED emitting materials are equally capable of it. Different approaches are needed because the luminescence mechanism and its complexity of calculation are different depending on the material types. For pure fluorescence or even high efficiency phosphorescence, their luminescence mechanisms are relatively well defined and nearly irreversible and so the correlation between the calculation and performance could be better. If so, the reverse design is becoming possible and already it has begun to be tried a lot. However, in the case of TADF, the radiation and non‐radiation paths vary, ISC‐RISC is more reversible, and the controversy over luminescence mechanism remains. As a result, the calculating method of luminous efficiency has not yet been fully established. In this study, we want to report the consistency level of predicting characteristics of OLED materials using AI, and also discuss the difference between each emitting material types for reverse design. In particular, we also want to share the issues of calculating methods for TADF performance.

We aim to separate the generative factors of data into two latent vectors in a variational autoencoder. One vector captures class factors relevant to target classification tasks, while the other vector captures style factors relevant to the remaining information. To learn the discrete class features, we introduce supervision using a small amount of labeled data, which can simply yet effectively reduce the effort required for hyperparameter tuning performed in existing unsupervised methods. Furthermore, we introduce a learning objective to encourage statistical independence between the vectors. We show that (i) this vector independence term exists within the result obtained on decomposing the evidence lower bound with multiple latent vectors, and (ii) encouraging such independence along with reducing the total correlation within the vectors enhances disentanglement performance. Experiments conducted on several image datasets demonstrate that the disentanglement achieved via our method can improve classification performance and generation controllability.

Voice conversion (VC) is a task to transform a person's voice to different style while conserving linguistic contents. Previous state-of-the-art on VC is based on sequence-to-sequence (seq2seq) model, which could mislead linguistic information. There was an attempt to overcome it by using textual supervision, it requires explicit alignment which loses the benefit of using seq2seq model. In this paper, a voice converter using multitask learning with text-to-speech (TTS) is presented. The embedding space of seq2seq-based TTS has abundant information on the text. The role of the decoder of TTS is to convert embedding space to speech, which is same to VC. In the proposed model, the whole network is trained to minimize loss of VC and TTS. VC is expected to capture more linguistic information and to preserve training stability by multitask learning. Experiments of VC were performed on a male Korean emotional text-speech dataset, and it is shown that multitask learning is helpful to keep linguistic contents in VC.

While electron-beam (e-beam) lithography is widely used in transferring fine-feature patterns onto a substrate, its major drawback is the low throughput, especially for large-scale patterns. To increase the writing throughput, e-beam machines with massively-parallel beams were recently developed. In such a system, it is highly likely that some beams may not be normal, e.g., permanently on or off, a significant current fluctuation, beam-positioning error, etc. Therefore, it is crucial to understand how abnormal beams affect writing qualities. In this study, the effects of abnormal beams on the writing qualities are analyzed through an extensive simulation, comparing three different writing methods, single-row writing I, single-row writing II, and multirow writing, to suggest ways of reducing their negative effects.

We propose an approach for digitally altering people's outfits in images. Given images of a person and a desired clothing style, our method generates a new clothing item image. The new item displays the color and pattern of the desired style while geometrically mimicking the person's original item. Through superimposition, the altered image is made to look as if the person is wearing the new item. Unlike recent works with full-image synthesis, our work relies on segment synthesis, yielding benefits in virtual try-on. For the synthesis process, we assume two underlying factors characterizing clothing segments: geometry and style. These two factors are disentangled via preprocessing and combined using a neural network. We explore several networks and introduce important aspects of the architecture and learning process. Our experimental results are three-fold: (1) on images from fashion-parsing datasets, we demonstrate the generation of high-quality clothing segments with fine-level style control; (2) on a virtual try-on benchmark, our method shows superiority over prior synthesis methods; (3) in transferring clothing styles, we visualize the differences between our method and neural style transfer.

Many speech enhancement methods try to learn the relationship between noisy and clean speech, obtained using an acoustic room simulator. We point out several limitations of enhancement methods relying on clean speech targets; the goal of this work is proposing an alternative learning algorithm, called acoustic and adversarial supervision (AAS). AAS makes the enhanced output both maximizing the likelihood of transcription on the pre-trained acoustic model and having general characteristics of clean speech, which improve generalization on unseen noisy speeches. We employ the connectionist temporal classification and the unpaired conditional boundary equilibrium generative adversarial network as the loss function of AAS. AAS is tested on two datasets including additive noise without and with reverberation, Librispeech + DEMAND and CHiME-4. By visualizing the enhanced speech with different loss combinations, we demonstrate the role of each supervision. AAS achieves a lower word error rate than other state-of-the-art methods using the clean speech target in both datasets.

Convolutional neural networks (CNNs) with convolutional and pooling operations along the frequency axis have been proposed to attain invariance to frequency shifts of features. However, this is inappropriate with regard to the fact that acoustic features vary in frequency. In this paper, we contend that convolution along the time axis is more effective. We also propose the addition of an intermap pooling (IMP) layer to deep CNNs. In this layer, filters in each group extract common but spectrally variant features, then the layer pools the feature maps of each group. As a result, the proposed IMP CNN can achieve insensitivity to spectral variations characteristic of different speakers and utterances. The effectiveness of the IMP CNN architecture is demonstrated on several LVCSR tasks. Even without speaker adaptation techniques, the architecture achieved a WER of 12.7% on the SWB part of the Hub5'2000 evaluation test set, which is competitive with other state-of-the-art methods.