Search

This paper introduces Neurocache, an approach to extend the effective context size of large language models (LLMs) using an external vector cache to store its past states. Like recent vector retrieval approaches, Neurocache uses an efficient k-nearest-neighbor (kNN) algorithm to retrieve relevant past states and incorporate them into the attention process. Neurocache improves upon previous methods by (1) storing compressed states, which reduces cache size; (2) performing a single retrieval operation per token which increases inference speed; and (3) extending the retrieval window to neighboring states, which improves both language modeling and downstream task accuracy. Our experiments show the effectiveness of Neurocache both for models trained from scratch and for pre-trained models such as Llama2-7B and Mistral-7B when enhanced with the cache mechanism. We also compare Neurocache with text retrieval methods and show improvements in single-document question-answering and few-shot learning tasks. We made the source code available under: https://github.com/alisafaya/neurocache, Comment: Long paper, published at the main conference NAACL'24

Large Language Models (LLMs) are becoming crucial across various fields, emphasizing the urgency for high-quality models in underrepresented languages. This study explores the unique challenges faced by low-resource languages, such as data scarcity, model selection, evaluation, and computational limitations, with a special focus on Turkish. We conduct an in-depth analysis to evaluate the impact of training strategies, model choices, and data availability on the performance of LLMs designed for underrepresented languages. Our approach includes two methodologies: (i) adapting existing LLMs originally pretrained in English to understand Turkish, and (ii) developing a model from the ground up using Turkish pretraining data, both supplemented with supervised fine-tuning on a novel Turkish instruction-tuning dataset aimed at enhancing reasoning capabilities. The relative performance of these methods is evaluated through the creation of a new leaderboard for Turkish LLMs, featuring benchmarks that assess different reasoning and knowledge skills. Furthermore, we conducted experiments on data and model scaling, both during pretraining and fine-tuning, simultaneously emphasizing the capacity for knowledge transfer across languages and addressing the challenges of catastrophic forgetting encountered during fine-tuning on a different language. Our goal is to offer a detailed guide for advancing the LLM framework in low-resource linguistic contexts, thereby making natural language processing (NLP) benefits more globally accessible.

The rise of large-scale multimodal models has paved the pathway for groundbreaking advances in generative modeling and reasoning, unlocking transformative applications in a variety of complex tasks. However, a pressing question that remains is their genuine capability for stronger forms of generalization, which has been largely underexplored in the multimodal setting. Our study aims to address this by examining sequential compositional generalization using \textsc{CompAct} (\underline{Comp}ositional \underline{Act}ivities)\footnote{Project Page: \url{http://cyberiada.github.io/CompAct}}, a carefully constructed, perceptually grounded dataset set within a rich backdrop of egocentric kitchen activity videos. Each instance in our dataset is represented with a combination of raw video footage, naturally occurring sound, and crowd-sourced step-by-step descriptions. More importantly, our setup ensures that the individual concepts are consistently distributed across training and evaluation sets, while their compositions are novel in the evaluation set. We conduct a comprehensive assessment of several unimodal and multimodal models. Our findings reveal that bi-modal and tri-modal models exhibit a clear edge over their text-only counterparts. This highlights the importance of multimodality while charting a trajectory for future research in this domain., Comment: Accepted to the main conference of NAACL (2024) as a long paper

Character re-identification, recognizing characters consistently across different panels in comics, presents significant challenges due to limited annotated data and complex variations in character appearances. To tackle this issue, we introduce a robust semi-supervised framework that combines metric learning with a novel 'Identity-Aware' self-supervision method by contrastive learning of face and body pairs of characters. Our approach involves processing both facial and bodily features within a unified network architecture, facilitating the extraction of identity-aligned character embeddings that capture individual identities while preserving the effectiveness of face and body features. This integrated character representation enhances feature extraction and improves character re-identification compared to re-identification by face or body independently, offering a parameter-efficient solution. By extensively validating our method using in-series and inter-series evaluation metrics, we demonstrate its effectiveness in consistently re-identifying comic characters. Compared to existing methods, our approach not only addresses the challenge of character re-identification but also serves as a foundation for downstream tasks since it can produce character embeddings without restrictions of face and body availability, enriching the comprehension of comic books. In our experiments, we leverage two newly curated datasets: the 'Comic Character Instances Dataset', comprising over a million character instances and the 'Comic Sequence Identity Dataset', containing annotations of identities within more than 3000 sets of four consecutive comic panels that we collected., Comment: 18 pages, 9 Figures

Researchers have recently begun exploring the use of StyleGAN-based models for real image editing. One particularly interesting application is using natural language descriptions to guide the editing process. Existing approaches for editing images using language either resort to instance-level latent code optimization or map predefined text prompts to some editing directions in the latent space. However, these approaches have inherent limitations. The former is not very efficient, while the latter often struggles to effectively handle multi-attribute changes. To address these weaknesses, we present CLIPInverter, a new text-driven image editing approach that is able to efficiently and reliably perform multi-attribute changes. The core of our method is the use of novel, lightweight text-conditioned adapter layers integrated into pretrained GAN-inversion networks. We demonstrate that by conditioning the initial inversion step on the CLIP embedding of the target description, we are able to obtain more successful edit directions. Additionally, we use a CLIP-guided refinement step to make corrections in the resulting residual latent codes, which further improves the alignment with the text prompt. Our method outperforms competing approaches in terms of manipulation accuracy and photo-realism on various domains including human faces, cats, and birds, as shown by our qualitative and quantitative results., Comment: Accepted for publication in ACM Transactions on Graphics

The algorithms used to train neural networks, like stochastic gradient descent (SGD), have close parallels to natural processes that navigate a high-dimensional parameter space -- for example protein folding or evolution. Our study uses a Fokker-Planck approach, adapted from statistical physics, to explore these parallels in a single, unified framework. We focus in particular on the stationary state of the system in the long-time limit, which in conventional SGD is out of equilibrium, exhibiting persistent currents in the space of network parameters. As in its physical analogues, the current is associated with an entropy production rate for any given training trajectory. The stationary distribution of these rates obeys the integral and detailed fluctuation theorems -- nonequilibrium generalizations of the second law of thermodynamics. We validate these relations in two numerical examples, a nonlinear regression network and MNIST digit classification. While the fluctuation theorems are universal, there are other aspects of the stationary state that are highly sensitive to the training details. Surprisingly, the effective loss landscape and diffusion matrix that determine the shape of the stationary distribution vary depending on the simple choice of minibatching done with or without replacement. We can take advantage of this nonequilibrium sensitivity to engineer an equilibrium stationary state for a particular application: sampling from a posterior distribution of network weights in Bayesian machine learning. We propose a new variation of stochastic gradient Langevin dynamics (SGLD) that harnesses without replacement minibatching. In an example system where the posterior is exactly known, this SGWORLD algorithm outperforms SGLD, converging to the posterior orders of magnitude faster as a function of the learning rate., Comment: 24 pages, 6 figures

This study focuses on improving the optical character recognition (OCR) data for panels in the COMICS dataset, the largest dataset containing text and images from comic books. To do this, we developed a pipeline for OCR processing and labeling of comic books and created the first text detection and recognition datasets for western comics, called "COMICS Text+: Detection" and "COMICS Text+: Recognition". We evaluated the performance of state-of-the-art text detection and recognition models on these datasets and found significant improvement in word accuracy and normalized edit distance compared to the text in COMICS. We also created a new dataset called "COMICS Text+", which contains the extracted text from the textboxes in the COMICS dataset. Using the improved text data of COMICS Text+ in the comics processing model from resulted in state-of-the-art performance on cloze-style tasks without changing the model architecture. The COMICS Text+ dataset can be a valuable resource for researchers working on tasks including text detection, recognition, and high-level processing of comics, such as narrative understanding, character relations, and story generation. All the data and inference instructions can be accessed in https://github.com/gsoykan/comics_text_plus., Comment: 33 pages, 10 figures, 16 tables

Drawings are powerful means of pictorial abstraction and communication. Understanding diverse forms of drawings, including digital arts, cartoons, and comics, has been a major problem of interest for the computer vision and computer graphics communities. Although there are large amounts of digitized drawings from comic books and cartoons, they contain vast stylistic variations, which necessitate expensive manual labeling for training domain-specific recognizers. In this work, we show how self-supervised learning, based on a teacher-student network with a modified student network update design, can be used to build face and body detectors. Our setup allows exploiting large amounts of unlabeled data from the target domain when labels are provided for only a small subset of it. We further demonstrate that style transfer can be incorporated into our learning pipeline to bootstrap detectors using a vast amount of out-of-domain labeled images from natural images (i.e., images from the real world). Our combined architecture yields detectors with state-of-the-art (SOTA) and near-SOTA performance using minimal annotation effort. Our code can be accessed from https://github.com/barisbatuhan/DASS_Detector., Comment: Preprint, 8 pages of the paper itself + 7 pages of Supplementary Material

This paper describes our winning systems in MRL: The 1st Shared Task on Multilingual Clause-level Morphology (EMNLP 2022 Workshop) designed by KUIS AI NLP team. We present our work for all three parts of the shared task: inflection, reinflection, and analysis. We mainly explore transformers with two approaches: (i) training models from scratch in combination with data augmentation, and (ii) transfer learning with prefix-tuning at multilingual morphological tasks. Data augmentation significantly improves performance for most languages in the inflection and reinflection tasks. On the other hand, Prefix-tuning on a pre-trained mGPT model helps us to adapt analysis tasks in low-data and multilingual settings. While transformer architectures with data augmentation achieved the most promising results for inflection and reinflection tasks, prefix-tuning on mGPT received the highest results for the analysis task. Our systems received 1st place in all three tasks in MRL 2022.

Self-supervised learning allows AI systems to learn effective representations from large amounts of data using tasks that do not require costly labeling. Mode collapse, i.e., the model producing identical representations for all inputs, is a central problem to many self-supervised learning approaches, making self-supervised tasks, such as matching distorted variants of the inputs, ineffective. In this article, we argue that a straightforward application of information maximization among alternative latent representations of the same input naturally solves the collapse problem and achieves competitive empirical results. We propose a self-supervised learning method, CorInfoMax, that uses a second-order statistics-based mutual information measure that reflects the level of correlation among its arguments. Maximizing this correlative information measure between alternative representations of the same input serves two purposes: (1) it avoids the collapse problem by generating feature vectors with non-degenerate covariances; (2) it establishes relevance among alternative representations by increasing the linear dependence among them. An approximation of the proposed information maximization objective simplifies to a Euclidean distance-based objective function regularized by the log-determinant of the feature covariance matrix. The regularization term acts as a natural barrier against feature space degeneracy. Consequently, beyond avoiding complete output collapse to a single point, the proposed approach also prevents dimensional collapse by encouraging the spread of information across the whole feature space. Numerical experiments demonstrate that CorInfoMax achieves better or competitive performance results relative to the state-of-the-art SSL approaches.

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting., Comment: 27 pages, 17 figures + references and appendices, repo: https://github.com/google/BIG-bench

Having sufficient resources for language X lifts it from the under-resourced languages class, but not necessarily from the under-researched class. In this paper, we address the problem of the absence of organized benchmarks in the Turkish language. We demonstrate that languages such as Turkish are left behind the state-of-the-art in NLP applications. As a solution, we present Mukayese, a set of NLP benchmarks for the Turkish language that contains several NLP tasks. We work on one or more datasets for each benchmark and present two or more baselines. Moreover, we present four new benchmarking datasets in Turkish for language modeling, sentence segmentation, and spell checking. All datasets and baselines are available under: https://github.com/alisafaya/mukayese, Comment: Accepted at Findings of ACL 2022 (Camera Ready)

Humans are able to perceive, understand and reason about causal events. Developing models with similar physical and causal understanding capabilities is a long-standing goal of artificial intelligence. As a step towards this direction, we introduce CRAFT, a new video question answering dataset that requires causal reasoning about physical forces and object interactions. It contains 58K video and question pairs that are generated from 10K videos from 20 different virtual environments, containing various objects in motion that interact with each other and the scene. Two question categories in CRAFT include previously studied descriptive and counterfactual questions. Additionally, inspired by the Force Dynamics Theory in cognitive linguistics, we introduce a new causal question category that involves understanding the causal interactions between objects through notions like cause, enable, and prevent. Our results show that even though the questions in CRAFT are easy for humans, the tested baseline models, including existing state-of-the-art methods, do not yet deal with the challenges posed in our benchmark., Comment: Accepted to Findings of ACL 2022

In this paper, we describe our approach to utilize pre-trained BERT models with Convolutional Neural Networks for sub-task A of the Multilingual Offensive Language Identification shared task (OffensEval 2020), which is a part of the SemEval 2020. We show that combining CNN with BERT is better than using BERT on its own, and we emphasize the importance of utilizing pre-trained language models for downstream tasks. Our system, ranked 4th with macro averaged F1-Score of 0.897 in Arabic, 4th with score of 0.843 in Greek, and 3rd with score of 0.814 in Turkish. Additionally, we present ArabicBERT, a set of pre-trained transformer language models for Arabic that we share with the community., Comment: to be published in the proceedings of the 14th International Workshop on Semantic Evaluation (SemEval2020), Association for Computational Linguistics (ACL)

How to best integrate linguistic and perceptual processing in multi-modal tasks that involve language and vision is an important open problem. In this work, we argue that the common practice of using language in a top-down manner, to direct visual attention over high-level visual features, may not be optimal. We hypothesize that the use of language to also condition the bottom-up processing from pixels to high-level features can provide benefits to the overall performance. To support our claim, we propose a U-Net-based model and perform experiments on two language-vision dense-prediction tasks: referring expression segmentation and language-guided image colorization. We compare results where either one or both of the top-down and bottom-up visual branches are conditioned on language. Our experiments reveal that using language to control the filters for bottom-up visual processing in addition to top-down attention leads to better results on both tasks and achieves competitive performance. Our linguistic analysis suggests that bottom-up conditioning improves segmentation of objects especially when input text refers to low-level visual concepts. Code is available at https://github.com/ilkerkesen/bvpr., Comment: 13 pages, 6 figures, 6 tables. Appeared in MULA Workshop at CVPR 2022

Human-robot interaction often occurs in the form of instructions given from a human to a robot. For a robot to successfully follow instructions, a common representation of the world and objects in it should be shared between humans and the robot so that the instructions can be grounded. Achieving this representation can be done via learning, where both the world representation and the language grounding are learned simultaneously. However, in robotics this can be a difficult task due to the cost and scarcity of data. In this paper, we tackle the problem by separately learning the world representation of the robot and the language grounding. While this approach can address the challenges in getting sufficient data, it may give rise to inconsistencies between both learned components. Therefore, we further propose Bayesian learning to resolve such inconsistencies between the natural language grounding and a robot's world representation by exploiting spatio-relational information that is implicitly present in instructions given by a human. Moreover, we demonstrate the feasibility of our approach on a scenario involving a robotic arm in the physical world.

In this paper, we present a state-of-the-art model and introduce a new dataset for grounded language learning. Our goal is to develop a model that can learn to follow new instructions given prior instruction-perception-action examples. We based our work on the SAIL dataset which consists of navigational instructions and actions in a maze-like environment. The new model we propose achieves the best results to date on the SAIL dataset by using an improved perceptual component that can represent relative positions of objects. We also analyze the problems with the SAIL dataset regarding its size and balance. We argue that performance on a small, fixed-size dataset is no longer a good measure to differentiate state-of-the-art models. We introduce SAILx, a synthetic dataset generator, and perform experiments where the size and balance of the dataset are controlled.

The encoder-decoder framework for neural machine translation (NMT) has been shown effective in large data scenarios, but is much less effective for low-resource languages. We present a transfer learning method that significantly improves Bleu scores across a range of low-resource languages. Our key idea is to first train a high-resource language pair (the parent model), then transfer some of the learned parameters to the low-resource pair (the child model) to initialize and constrain training. Using our transfer learning method we improve baseline NMT models by an average of 5.6 Bleu on four low-resource language pairs. Ensembling and unknown word replacement add another 2 Bleu which brings the NMT performance on low-resource machine translation close to a strong syntax based machine translation (SBMT) system, exceeding its performance on one language pair. Additionally, using the transfer learning model for re-scoring, we can improve the SBMT system by an average of 1.3 Bleu, improving the state-of-the-art on low-resource machine translation., Comment: 8 pages

We analyze a word embedding method in supervised tasks. It maps words on a sphere such that words co-occurring in similar contexts lie closely. The similarity of contexts is measured by the distribution of substitutes that can fill them. We compared word embeddings, including more recent representations, in Named Entity Recognition (NER), Chunking, and Dependency Parsing. We examine our framework in multilingual dependency parsing as well. The results show that the proposed method achieves as good as or better results compared to the other word embeddings in the tasks we investigate. It achieves state-of-the-art results in multilingual dependency parsing. Word embeddings in 7 languages are available for public use., Comment: 11 pages

Relevance of mode coupling to energy/information transfer during protein function, particularly in the context of allosteric interactions is widely accepted. However, existing evidence in favor of this hypothesis comes essentially from model systems. We here report a novel formal analysis of the near-native dynamics of myosin II, which allows us to explore the impact of the interaction between possibly non-Gaussian vibrational modes on fluctutational dynamics. We show that, an information-theoretic measure based on mode coupling {\it alone} yields a ranking of residues with a statistically significant bias favoring the functionally critical locations identified by experiments on myosin II., Comment: 17 pages, 6 figures, 1 table

Lexical substitutes have found use in areas such as paraphrasing, text simplification, machine translation, word sense disambiguation, and part of speech induction. However the computational complexity of accurately identifying the most likely substitutes for a word has made large scale experiments difficult. In this paper I introduce a new search algorithm, FASTSUBS, that is guaranteed to find the K most likely lexical substitutes for a given word in a sentence based on an n-gram language model. The computation is sub-linear in both K and the vocabulary size V. An implementation of the algorithm and a dataset with the top 100 substitutes of each token in the WSJ section of the Penn Treebank are available at http://goo.gl/jzKH0., Comment: 4 pages, 1 figure, to appear in IEEE Signal Processing Letters

This work has been motivated by two long term goals: to understand how humans learn language and to build programs that can understand language. Using a representation that makes the relevant features explicit is a prerequisite for successful learning and understanding. Therefore, I chose to represent relations between individual words explicitly in my model. Lexical attraction is defined as the likelihood of such relations. I introduce a new class of probabilistic language models named lexical attraction models which can represent long distance relations between words and I formalize this new class of models using information theory. Within the framework of lexical attraction, I developed an unsupervised language acquisition program that learns to identify linguistic relations in a given sentence. The only explicitly represented linguistic knowledge in the program is lexical attraction. There is no initial grammar or lexicon built in and the only input is raw text. Learning and processing are interdigitated. The processor uses the regularities detected by the learner to impose structure on the input. This structure enables the learner to detect higher level regularities. Using this bootstrapping procedure, the program was trained on 100 million words of Associated Press material and was able to achieve 60% precision and 50% recall in finding relations between content-words. Using knowledge of lexical attraction, the program can identify the correct relations in syntactically ambiguous sentences such as ``I saw the Statue of Liberty flying over New York.'', Comment: dissertation, 56 pages