Several popular language models represent local contexts in an input text as bags of words. Such representations are naturally encoded by a sequence graph whose vertices are the distinct words occurring in x, with edges representing the (ordered) co-occurrence of two words within a sliding window of size w. However, this compressed representation is not generally bijective, and may introduce some degree of ambiguity. Some sequence graphs may admit several realizations as a sequence, while others may not admit any realization. In this paper, we study the realizability and ambiguity of sequence graphs from a combinatorial and computational point of view. We consider the existence and enumeration of realizations of a sequence graph under multiple settings: window size w, presence/absence of graph orientation, and presence/absence of weights (multiplicities). When w = 2, we provide polynomial time algorithms for realizability and enumeration in all cases except the undirected/weighted setting, where we show the #P-hardness of enumeration. For a window of size at least 3, we prove hardness of all variants, even when w is considered as a constant, with the notable exception of the undirected/unweighted case for which we propose an XP algorithms for both (realizability and enumeration) problems, tight due to a corresponding W[1]-hardness result. We conclude with an integer program formulation to solve the realizability problem, and with dynamic programming to solve the enumeration problem. This work leaves open the membership to NP for both problems, a non-trivial question due to the existence of minimum realizations having exponential size on the instance encoding.
相關內容
Large language models (LLMs) often generate content that contains factual errors when responding to fact-seeking prompts on open-ended topics. To benchmark a model's long-form factuality in open domains, we first use GPT-4 to generate LongFact, a prompt set comprising thousands of questions spanning 38 topics. We then propose that LLM agents can be used as automated evaluators for long-form factuality through a method which we call Search-Augmented Factuality Evaluator (SAFE). SAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results. Furthermore, we propose extending F1 score as an aggregated metric for long-form factuality. To do so, we balance the percentage of supported facts in a response (precision) with the percentage of provided facts relative to a hyperparameter representing a user's preferred response length (recall). Empirically, we demonstrate that LLM agents can achieve superhuman rating performance - on a set of ~16k individual facts, SAFE agrees with crowdsourced human annotators 72% of the time, and on a random subset of 100 disagreement cases, SAFE wins 76% of the time. At the same time, SAFE is more than 20 times cheaper than human annotators. We also benchmark thirteen language models on LongFact across four model families (Gemini, GPT, Claude, and PaLM-2), finding that larger language models generally achieve better long-form factuality. LongFact, SAFE, and all experimental code are available at //github.com/google-deepmind/long-form-factuality.
Languages are grouped into families that share common linguistic traits. While this approach has been successful in understanding genetic relations between diverse languages, more analyses are needed to accurately quantify their relatedness, especially in less studied linguistic levels such as syntax. Here, we explore linguistic distances using series of parts of speech (POS) extracted from the Universal Dependencies dataset. Within an information-theoretic framework, we show that employing POS trigrams maximizes the possibility of capturing syntactic variations while being at the same time compatible with the amount of available data. Linguistic connections are then established by assessing pairwise distances based on the POS distributions. Intriguingly, our analysis reveals definite clusters that correspond to well known language families and groups, with exceptions explained by distinct morphological typologies. Furthermore, we obtain a significant correlation between language similarity and geographic distance, which underscores the influence of spatial proximity on language kinships.
State-of-the-art language models (LMs) sometimes generate non-factual hallucinations that misalign with world knowledge. Despite extensive efforts to detect and mitigate hallucinations, understanding their internal mechanisms remains elusive. Our study investigates the mechanistic causes of hallucination, specifically non-factual ones where the LM incorrectly predicts object attributes in response to subject-relation queries. With causal mediation analysis and embedding space projection, we identify two general mechanistic causes of hallucinations shared across LMs of various scales and designs: 1) insufficient subject attribute knowledge in lower layer MLPs, and 2) failing to select the correct object attribute in upper layer attention heads and MLPs. These two mechanisms exhibit varying degrees of subject-object association, predictive uncertainty and perturbation robustness. Additionally, we scrutinize LM pre-training checkpoints, revealing distinct learning dynamics for the two mechanistic causes of hallucinations. We also highlight how attribution features from our causal analysis can effectively construct hallucination detectors. Our work proposes a mechanistic understanding of LM factual errors.
Antibodies are proteins produced by the immune system that can identify and neutralise a wide variety of antigens with high specificity and affinity, and constitute the most successful class of biotherapeutics. With the advent of next-generation sequencing, billions of antibody sequences have been collected in recent years, though their application in the design of better therapeutics has been constrained by the sheer volume and complexity of the data. To address this challenge, we present IgBert and IgT5, the best performing antibody-specific language models developed to date which can consistently handle both paired and unpaired variable region sequences as input. These models are trained comprehensively using the more than two billion unpaired sequences and two million paired sequences of light and heavy chains present in the Observed Antibody Space dataset. We show that our models outperform existing antibody and protein language models on a diverse range of design and regression tasks relevant to antibody engineering. This advancement marks a significant leap forward in leveraging machine learning, large scale data sets and high-performance computing for enhancing antibody design for therapeutic development.
We study infinite binary words that contain few distinct palindromes. In particular, we classify such words according to their critical exponents. This extends results by Fici and Zamboni [TCS 2013]. Interestingly, the words with 18 and 20 palindromes happen to be morphic images of the fixed point of the morphism $\texttt{0}\mapsto\texttt{01}$, $\texttt{1}\mapsto\texttt{21}$, $\texttt{2}\mapsto\texttt{0}$.
The advent of ChatGPT and similar large language models (LLMs) has revolutionized the human-AI interaction and information-seeking process. Leveraging LLMs as an alternative to search engines, users can now access summarized information tailored to their queries, significantly reducing the cognitive load associated with navigating vast information resources. This shift underscores the potential of LLMs in redefining information access paradigms. Drawing on the foundation of task-focused information retrieval and LLMs' task planning ability, this research extends the scope of LLM capabilities beyond routine task automation to support users in navigating long-term and significant life tasks. It introduces the GOLF framework (Goal-Oriented Long-term liFe tasks), which focuses on enhancing LLMs' ability to assist in significant life decisions through goal orientation and long-term planning. The methodology encompasses a comprehensive simulation study to test the framework's efficacy, followed by model and human evaluations to develop a dataset benchmark for long-term life tasks, and experiments across different models and settings. By shifting the focus from short-term tasks to the broader spectrum of long-term life goals, this research underscores the transformative potential of LLMs in enhancing human decision-making processes and task management, marking a significant step forward in the evolution of human-AI collaboration.
This paper makes two contributions to the field of text-based patent similarity. First, it compares the performance of different kinds of patent-specific pretrained embedding models, namely static word embeddings (such as word2vec and doc2vec models) and contextual word embeddings (such as transformers based models), on the task of patent similarity calculation. Second, it compares specifically the performance of Sentence Transformers (SBERT) architectures with different training phases on the patent similarity task. To assess the models' performance, we use information about patent interferences, a phenomenon in which two or more patent claims belonging to different patent applications are proven to be overlapping by patent examiners. Therefore, we use these interferences cases as a proxy for maximum similarity between two patents, treating them as ground-truth to evaluate the performance of the different embedding models. Our results point out that, first, Patent SBERT-adapt-ub, the domain adaptation of the pretrained Sentence Transformer architecture proposed in this research, outperforms the current state-of-the-art in patent similarity. Second, they show that, in some cases, large static models performances are still comparable to contextual ones when trained on extensive data; thus, we believe that the superiority in the performance of contextual embeddings may not be related to the actual architecture but rather to the way the training phase is performed.
The front-end is a critical component of English text-to-speech (TTS) systems, responsible for extracting linguistic features that are essential for a text-to-speech model to synthesize speech, such as prosodies and phonemes. The English TTS front-end typically consists of a text normalization (TN) module, a prosody word prosody phrase (PWPP) module, and a grapheme-to-phoneme (G2P) module. However, current research on the English TTS front-end focuses solely on individual modules, neglecting the interdependence between them and resulting in sub-optimal performance for each module. Therefore, this paper proposes a unified front-end framework that captures the dependencies among the English TTS front-end modules. Extensive experiments have demonstrated that the proposed method achieves state-of-the-art (SOTA) performance in all modules.
In the search for highly efficient decoders for short LDPC codes approaching maximum likelihood performance, a relayed decoding strategy, specifically activating the ordered statistics decoding process upon failure of a neural min-sum decoder, is enhanced by instilling three innovations. Firstly, soft information gathered at each step of the neural min-sum decoder is leveraged to forge a new reliability measure using a convolutional neural network. This measure aids in constructing the most reliable basis of ordered statistics decoding, bolstering the decoding process by excluding error-prone bits or concentrating them in a smaller area. Secondly, an adaptive ordered statistics decoding process is introduced, guided by a derived decoding path comprising prioritized blocks, each containing distinct test error patterns. The priority of these blocks is determined from the statistical data during the query phase. Furthermore, effective complexity management methods are devised by adjusting the decoding path's length or refining constraints on the involved blocks. Thirdly, a simple auxiliary criterion is introduced to reduce computational complexity by minimizing the number of candidate codewords before selecting the optimal estimate. Extensive experimental results and complexity analysis strongly support the proposed framework, demonstrating its advantages in terms of high throughput, low complexity, independence from noise variance, in addition to superior decoding performance.
Coverage of ChatGPT-style large language models (LLMs) in the media has focused on their eye-catching achievements, including solving advanced mathematical problems and reaching expert proficiency in medical examinations. But the gradual adoption of LLMs in agriculture, an industry which touches every human life, has received much less public scrutiny. In this short perspective, we examine risks and opportunities related to more widespread adoption of language models in food production systems. While LLMs can potentially enhance agricultural efficiency, drive innovation, and inform better policies, challenges like agricultural misinformation, collection of vast amounts of farmer data, and threats to agricultural jobs are important concerns. The rapid evolution of the LLM landscape underscores the need for agricultural policymakers to think carefully about frameworks and guidelines that ensure the responsible use of LLMs in food production before these technologies become so ingrained that policy intervention becomes challenging.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191