Representation learning for proteins has primarily focused on the global understanding of protein sequences regardless of their length. However, shorter proteins (known as peptides) take on distinct structures and functions compared to their longer counterparts. Unfortunately, there are not as many naturally occurring peptides available to be sequenced and therefore less peptide-specific data to train with. In this paper, we propose a new peptide data augmentation scheme, where we train peptide language models on artificially constructed peptides that are small contiguous subsets of longer, wild-type proteins; we refer to the training peptides as "chopped proteins". We evaluate the representation potential of models trained with chopped proteins versus natural peptides and find that training language models with chopped proteins results in more generalized embeddings for short protein sequences. These peptide-specific models also retain information about the original protein they were derived from better than language models trained on full-length proteins. We compare masked language model training objectives to three novel peptide-specific training objectives: next-peptide prediction, contrastive peptide selection and evolution-weighted MLM. We demonstrate improved zero-shot learning performance for a deep mutational scan peptides benchmark.

A major concern of Machine Learning (ML) models is their opacity. They are deployed in an increasing number of applications where they often operate as black boxes that do not provide explanations for their predictions. Among others, the potential harms associated with the lack of understanding of the models' rationales include privacy violations, adversarial manipulations, and unfair discrimination. As a result, the accountability and transparency of ML models have been posed as critical desiderata by works in policy and law, philosophy, and computer science. In computer science, the decision-making process of ML models has been studied by developing accountability and transparency methods. Accountability methods, such as adversarial attacks and diagnostic datasets, expose vulnerabilities of ML models that could lead to malicious manipulations or systematic faults in their predictions. Transparency methods explain the rationales behind models' predictions gaining the trust of relevant stakeholders and potentially uncovering mistakes and unfairness in models' decisions. To this end, transparency methods have to meet accountability requirements as well, e.g., being robust and faithful to the underlying rationales of a model. This thesis presents my research that expands our collective knowledge in the areas of accountability and transparency of ML models developed for complex reasoning tasks over text.

Large language models (LLMs) have led to a series of breakthroughs in natural language processing (NLP), owing to their excellent understanding and generation abilities. Remarkably, what further sets these models apart is the massive amounts of world knowledge they internalize during pretraining. While many downstream applications provide the model with an informational context to aid its performance on the underlying task, how the model's world knowledge interacts with the factual information presented in the context remains under explored. As a desirable behavior, an LLM should give precedence to the context whenever it contains task-relevant information that conflicts with the model's memorized knowledge. This enables model predictions to be grounded in the context, which can then be used to update or correct specific model predictions without frequent retraining. By contrast, when the context is irrelevant to the task, the model should ignore it and fall back on its internal knowledge. In this paper, we undertake a first joint study of the aforementioned two properties, namely controllability and robustness, in the context of LLMs. We demonstrate that state-of-the-art T5 and PaLM (both pretrained and finetuned) could exhibit poor controllability and robustness, which do not scale with increasing model size. As a solution, we propose a novel method - Knowledge Aware FineTuning (KAFT) - to strengthen both controllability and robustness by incorporating counterfactual and irrelevant contexts to standard supervised datasets. Our comprehensive evaluation showcases the utility of KAFT across model architectures and sizes.

We study the problem of efficient generative inference for Transformer models, in one of its most challenging settings: large deep models, with tight latency targets and long sequence lengths. Better understanding of the engineering tradeoffs for inference for large Transformer-based models is important as use cases of these models are growing rapidly throughout application areas. We develop a simple analytical model for inference efficiency to select the best multi-dimensional partitioning techniques optimized for TPU v4 slices based on the application requirements. We combine these with a suite of low-level optimizations to achieve a new Pareto frontier on the latency and model FLOPS utilization (MFU) tradeoffs on 500B+ parameter models that outperforms the FasterTransformer suite of benchmarks. We further show that with appropriate partitioning, the lower memory requirements of multiquery attention (i.e. multiple query heads share single key/value head) enables scaling up to 32x larger context lengths. Finally, we achieve a low-batch-size latency of 29ms per token during generation (using int8 weight quantization) and a 76% MFU during large-batch-size processing of input tokens, while supporting a long 2048-token context length on the PaLM 540B parameter model.

Misinformation spread over social media has become an undeniable infodemic. However, not all spreading claims are made equal. If propagated, some claims can be destructive, not only on the individual level, but to organizations and even countries. Detecting claims that should be prioritized for fact-checking is considered the first step to fight against spread of fake news. With training data limited to a handful of languages, developing supervised models to tackle the problem over lower-resource languages is currently infeasible. Therefore, our work aims to investigate whether we can use existing datasets to train models for predicting worthiness of verification of claims in tweets in other languages. We present a systematic comparative study of six approaches for cross-lingual check-worthiness estimation across pairs of five diverse languages with the help of Multilingual BERT (mBERT) model. We run our experiments using a state-of-the-art multilingual Twitter dataset. Our results show that for some language pairs, zero-shot cross-lingual transfer is possible and can perform as good as monolingual models that are trained on the target language. We also show that in some languages, this approach outperforms (or at least is comparable to) state-of-the-art models.

Precisely assessing the progress in natural language generation (NLG) tasks is challenging, and human evaluation to establish a preference in a model's output over another is often necessary. However, human evaluation is usually costly, difficult to reproduce, and non-reusable. In this paper, we propose a new and simple automatic evaluation method for NLG called Near-Negative Distinction (NND) that repurposes prior human annotations into NND tests. In an NND test, an NLG model must place a higher likelihood on a high-quality output candidate than on a near-negative candidate with a known error. Model performance is established by the number of NND tests a model passes, as well as the distribution over task-specific errors the model fails on. Through experiments on three NLG tasks (question generation, question answering, and summarization), we show that NND achieves a higher correlation with human judgments than standard NLG evaluation metrics. We then illustrate NND evaluation in four practical scenarios, for example performing fine-grain model analysis, or studying model training dynamics. Our findings suggest that NND can give a second life to human annotations and provide low-cost NLG evaluation.

Microsoft and its computer code-sharing website GitHub, as well as artificial intelligence firm OpenAI, are being sued in California. "We contend that the defendants have violated the legal rights of a vast number of creators who posted code or other work …

Generating text with autoregressive language models (LMs) is of great importance to many natural language processing (NLP) applications. Previous solutions for this task often produce text that contains degenerative expressions or lacks semantic consistency. Recently, Su et al. introduced a new decoding method, contrastive search, based on the isotropic representation space of the language model and obtained new state of the art on various benchmarks. Additionally, Su et al. argued that the representations of autoregressive LMs (e.g. GPT-2) are intrinsically anisotropic which is also shared by previous study.

A new AI revolution is upon us, and it is no longer a secret. In a matter of only a few months, recent LLM (large language model) breakthroughs have propelled AI from a niche trend to the dominating force behind much of the tech startup landscape. However, unlike other recent hype trends of the past few years, the value of AI has quickly become clear, showing us all how photos, writing, video, and other forms of creative media will vastly evolve in the coming years. Generative AI has the potential to transform content creation and consumption across industries including social media, gaming, immersive reality, entertainment, commerce, among many other consumer categories. However, while much of the world's attention has been focused on AI's impact on creative media in recent months, one of the earliest use cases of AI stands to also greatly benefit from recent tech advancements: AI-powered bots, or non-playable characters (NPCs).

Natural language prompts have been shown to facilitate cross-task generalization for large language models. However, with no or limited labeled examples, the cross-task performance is highly sensitive to the choice of prompts, while selecting a high-performing prompt is challenging given the scarcity of labels. To address the issue, we propose a Zero-Label Prompt Selection (ZPS) method that selects prompts without any labeled data or gradient update. Specifically, given the candidate human-written prompts for a task, ZPS labels a set of unlabeled data with a prompt ensemble and uses the pseudo-labels for prompt selection. Experiments show that ZPS improves over prior methods by a sizeable margin in zero-label performance. We also extend ZPS to a few-shot setting and show its advantages over strong baselines such as prompt tuning and model tuning. Recently, extensive studies have shown that large language models (LLMs) have promising performance for few-shot learning (Brown et al., 2020; Zhao et al., 2021; Schick & Schütze, 2021; Gao et al., 2021), and they even show strong generalization abilities to new tasks without any annotated data (Brown et al., 2020; Wei et al., 2021; Sanh et al., 2021). Different from conventional fine-tuning methods that require expensive parameter updates for each downstream task, prompts are employed to provide in-context information or task instructions, which is helpful for guiding models to perform each task. Manually-written prompts are often used to specify the task and unify the format of inputs. However, the performance of different prompts during evaluation can vary from near state-of-the-art to random guess; e.g., using a non-optimal prompt can cause a performance drop of up to 60 points on the CB task (Zhao et al., 2021).