We introduce AdamS, a simple yet effective alternative to Adam for large language model (LLM) pretraining and post-training.By leveraging a novel denominator, i.e., the root of weighted sum of squares of the momentum and the current gradient, AdamS eliminates the need for second-moment estimates.Hence, AdamS is efficient, matching the memory and compute footprint of SGD with momentum while delivering superior optimization performance.Moreover, AdamS is easy to adopt: it can directly inherit hyperparameters of AdamW, and is entirely model-agnostic, integrating seamlessly into existing pipelines without modifications to optimizer APIs or architectures.The motivation behind AdamS stems from the observed (L 0 , L 1 ) smoothness properties in transformer objectives, where local smoothness is governed by gradient magnitudes that can be further approximated by momentum magnitudes.We establish rigorous theoretical convergence guarantees and provide practical guidelines for hyperparameter selection.Empirically, AdamS demonstrates strong performance in various tasks, including pre-training runs on GPT-2 and Llama2 (up to 13B parameters) and reinforcement learning in post-training regimes.With its efficiency, simplicity, and theoretical grounding, AdamS stands as a compelling alternative to existing optimizers.

We present results from a large-scale experiment on pretraining encoders with non-embedding parameter counts ranging from 700M to 9.3B, their subsequent distillation into smaller models ranging from 17M-170M parameters, and their application to the Natural Language Understanding (NLU) component of a virtual assistant system. Though we train using 70% spoken-form data, our teacher models perform comparably to XLM-R and mT5 when evaluated on the written-form Cross-lingual Natural Language Inference (XNLI) corpus. We perform a second stage of pretraining on our teacher models using in-domain data from our system, improving error rates by 3.86% relative for intent classification and 7.01% relative for slot filling. We find that even a 170M-parameter model distilled from our Stage 2 teacher model has 2.88% better intent classification and 7.69% better slot filling error rates when compared to the 2.3B-parameter teacher trained only on public data (Stage 1), emphasizing the importance of in-domain data for pretraining. When evaluated offline using labeled NLU data, our 17M-parameter Stage 2 distilled model outperforms both XLM-R Base (85M params) and DistillBERT (42M params) by 4.23% to 6.14%, respectively. Finally, we present results from a full virtual assistant experimentation platform, where we find that models trained using our pretraining and distillation pipeline outperform models distilled from 85M-parameter teachers by 3.74%-4.91% on an automatic measurement of full-system user dissatisfaction.

Current spoken language understanding systems heavily rely on the best hypothesis (ASR 1-best) generated by automatic speech recognition, which is used as the input for downstream models such as natural language understanding (NLU) modules. However, the potential errors and misrecognition in ASR 1-best raise challenges to NLU. It is usually difficult for NLU models to recover from ASR errors without additional signals, which leads to suboptimal SLU performance. This paper proposes a fusion network to jointly consider ASR n-best hypotheses for enhanced robustness to ASR errors. Our experiments on Alexa data show that our model achieved 21.71% error reduction compared to baseline trained on transcription for domain classification.

Virtual adversarial training (VAT) is a powerful technique to improve model robustness in both supervised and semi-supervised settings. It is effective and can be easily adopted on lots of image classification and text classification tasks. However, its benefits to sequence labeling tasks such as named entity recognition (NER) have not been shown as significant, mostly, because the previous approach can not combine VAT with the conditional random field (CRF). CRF can significantly boost accuracy for sequence models by putting constraints on label transitions, which makes it an essential component in most state-of-the-art sequence labeling model architectures. In this paper, we propose SeqVAT, a method which naturally applies VAT to sequence labeling models with CRF. Empirical studies show that SeqVAT not only significantly improves the sequence labeling performance over baselines under supervised settings, but also outperforms state-of-the-art approaches under semi-supervised settings.

*: Background Consisting of dictated free-text documents such as discharge summaries, medical narratives are widely used in medical natural language processing. Relationships between anatomical entities and human body parts are crucial for building medical text mining applications. To achieve this, we establish a mapping system consisting of a Wikipedia-based scoring algorithm and a named entity normalization method (NEN). The mapping system makes full use of information available on Wikipedia, which is a comprehensive Internet medical knowledge base. We also built a new ontology, Tree of Human Body Parts (THBP), from core anatomical parts by referring to anatomical experts and Unified Medical Language Systems (UMLS) to make the mapping system efficacious for clinical treatments. *: Result The gold standard is derived from 50 discharge summaries from our previous work, in which 2,224 anatomical entities are included. The F1-measure of the baseline system is 70.20%, while our algorithm based on Wikipedia achieves 86.67% with the assistance of NEN. *: Conclusions We construct a framework to map anatomical entities to THBP ontology using normalization and a scoring algorithm based on Wikipedia. The proposed framework is proven to be much more effective and efficient than the main baseline system.

BACKGROUND: Electronic medical record (EMR) systems have become widely used throughout the world to improve the quality of healthcare and the efficiency of hospital services. A bilingual medical lexicon of Chinese and English is needed to meet the demand for the multi-lingual and multi-national treatment. We make efforts to extract a bilingual lexicon from English and Chinese discharge summaries with a small seed lexicon. The lexical terms can be classified into two categories: single-word terms (SWTs) and multi-word terms (MWTs). For SWTs, we use a label propagation (LP; context-based) method to extract candidates of translation pairs. For MWTs, which are pervasive in the medical domain, we propose a term alignment method, which firstly obtains translation candidates for each component word of a Chinese MWT, and then generates their combinations, from which the system selects a set of plausible translation candidates. RESULTS: We compare our LP method with a baseline method based on simple context-similarity. The LP based method outperforms the baseline with the accuracies: 4.44% Acc1, 24.44% Acc10, and 62.22% Acc100, where AccN means the top N accuracy. The accuracy of the LP method drops to 5.41% Acc10 and 8.11% Acc20 for MWTs. Our experiments show that the method based on term alignment improves the performance for MWTs to 16.22% Acc10 and 27.03% Acc20. CONCLUSIONS: We constructed a framework for building an English-Chinese term dictionary from discharge summaries in the two languages. Our experiments have shown that the LP-based method augmented with the term alignment method will contribute to reduction of manual work required to compile a bilingual sydictionary of clinical terms.