Research Blog

ASAPP - To measure the performance of Conversational AI, we need more strict, better quality benchmarks

To measure the performance of Conversational AI, we need more strict, better quality benchmarks

For conversational AI to advance, the broader scientific community must be able to work together and explore with easily accessible state-of-the-art baselines for fair performance comparisons. Read more

Suwon Shon
Suwon Shon, PhD

Senior Speech Scientist at ASAPP

ASAPP - To measure the performance of Conversational AI, we need more strict, better quality benchmarks
ASAPP - Wav2vec could be more efficient. So we created our own pre-trained ASR Model for better Conversational AI

Wav2vec could be more efficient, so we created our own pre-trained ASR Model for better Conversational AI.

Wav2vec 2.0 is arguably the most popular approach for using self-supervised training in speech, but it could be more efficient. We introduce SEW for better efficiency and performance. Read more

Felix Wu
Felix Wu, PhD

Research Scientist at ASAPP

ASAPP - Wav2vec could be more efficient. So we created our own pre-trained ASR Model for better Conversational AI
ASAPP - GOLD: Improving Out-of-Scope Detection in Dialogues using Data Augmentation

GOLD: Improving Out-of-Scope Detection in Dialogues using Data Augmentation

New ASAPP research introduces an orthogonal technique that augments existing data to train better out of scope detectors operating in low-data regimes. Read more

Derek Chen
Derek Chen

Research Scientist at ASAPP

ASAPP - GOLD: Improving Out-of-Scope Detection in Dialogues using Data Augmentation
ASAPP—Utilizing Pre-trained Language Model for Speech Sentiment Analysis

Utilizing Pre-trained Language Model for Speech Sentiment Analysis

On the path to real-time speech sentiment analysis, new ASAPP research achieves training efficiency gains with transfer learning between spoken and written language domains. Read more

Suwon Shon
Suwon Shon, PhD

Senior Speech Scientist at ASAPP

ASAPP—Utilizing Pre-trained Language Model for Speech Sentiment Analysis
ASAPP—Multi-mode ASR: Increasing Robustness with Dynamic Future Contexts

Multi-mode ASR: Increasing Robustness with Dynamic Future Contexts

Rather than maintaining multiple ASR models that work under varying time constraints or conditions, new ASAPP research introduces a single multi-mode model that can dynamically adjust to different scenarios. Read more

Kwangyoun Kim
Kwangyoun Kim

Senior Speech Scientist at ASAPP

ASAPP—Multi-mode ASR: Increasing Robustness with Dynamic Future Contexts
ASAPP—Introducing CLIP: A Dataset to Improve Continuity of Patient Care with Unsupervised NLP

Introducing CLIP: A Dataset to Improve Continuity of Patient Care with Unsupervised NLP

In pursuit of our mission to enhance human performance and automate the world’s workflows, ASAPP is releasing one of the largest annotated datasets for clinical NLP. Read more

James Mullenbach
James Mullenbach

Research Engineer at ASAPP

ASAPP—Introducing CLIP: A Dataset to Improve Continuity of Patient Care with Unsupervised NLP
ASAPP—Task-oriented dialogue systems could be better. Here’s a new dataset to help.

Task-oriented dialogue systems could be better. Here’s a new dataset to help.

Dialogue State Tracking has run its course. That’s why we’re establishing a new Action-Based Conversations Dataset. Read more

Derek Chen
Derek Chen

Research Scientist at ASAPP

ASAPP—Task-oriented dialogue systems could be better. Here’s a new dataset to help.
ASAPP—Addressing Instabilities For Few-sample BERT Fine-tuning

Addressing instabilities for few-sample BERT fine-tuning

Building on recent advances in natural language processing, new research from Felix Wu identifies ways to significantly stabilize BERT fine-tuning on small datasets. Read more

Felix Wu
Felix Wu, PhD

Research Scientist at ASAPP

ASAPP—Addressing Instabilities For Few-sample BERT Fine-tuning
ASAPP—Filling in the missing pieces for automation

Filling in the missing pieces for automation

Natural language input can be hard to classify. ASAPP research goes beyond conventional methods, building better systems to inform more accurate automation. Read more

Yoav Artzi
Yoav Artzi

Research Fellow at ASAPP

ASAPP—Filling in the missing pieces for automation
ASAPP—From Network Compression to DenseNets

From network compression to DenseNets

Neural Network Compression. What is it, and why does it matter? Here’s a look at what led to the development of DenseNets for parameter-efficient networks with significantly more accurate predictions. Read more

Kilian Weinberger
Kilian Weinberger, PhD

Principal Scientist and Head of ASAPP Ithaca Research Lab

ASAPP—From Network Compression to DenseNets