Alon Jacovi

Alon Jacovi, Moran Ambar, Eyal Ben-David et al. · deepmind

There is a growing line of research on verifying the correctness of language models' outputs. At the same time, LMs are being used to tackle complex queries that require reasoning. We introduce CoverBench, a challenging benchmark focused on verifying LM outputs in complex reasoning settings. Datasets that can be used for this purpose are often designed for other complex reasoning tasks (e.g., QA) targeting specific use-cases (e.g., financial tables), requiring transformations, negative sampling and selection of hard examples to collect such a benchmark. CoverBench provides a diversified evaluation for complex claim verification in a variety of domains, types of reasoning, relatively long inputs, and a variety of standardizations, such as multiple representations for tables where available, and a consistent schema. We manually vet the data for quality to ensure low levels of label noise. Finally, we report a variety of competitive baseline results to show CoverBench is challenging and has very significant headroom. The data is available at https://huggingface.co/datasets/google/coverbench .

Alon Jacovi, Avi Caciularu, Jonathan Herzig et al. · deepmind

A growing area of research investigates augmenting language models with tools (e.g., search engines, calculators) to overcome their shortcomings (e.g., missing or incorrect knowledge, incorrect logical inferences). Various few-shot tool-usage strategies have been proposed. However, there is no systematic and fair comparison across different strategies, or between these strategies and strong baselines that do not leverage tools. We conduct an extensive empirical analysis, finding that (1) across various datasets, example difficulty levels, and models, strong no-tool baselines are competitive to tool-assisted strategies, implying that effectively using tools with in-context demonstrations is a difficult unsolved problem; (2) for knowledge-retrieval tasks, strategies that *refine* incorrect outputs with tools outperform strategies that retrieve relevant information *ahead of* or *during generation*; (3) tool-assisted strategies are expensive in the number of tokens they require to work -- incurring additional costs by orders of magnitude -- which does not translate into significant improvement in performance. Overall, our findings suggest that few-shot tool integration is still an open challenge, emphasizing the need for comprehensive evaluations of future strategies to accurately assess their *benefits* and *costs*.

Oscar Sainz, Iker García-Ferrero, Alon Jacovi et al. · ibm-research

The 1st Workshop on Data Contamination (CONDA 2024) focuses on all relevant aspects of data contamination in natural language processing, where data contamination is understood as situations where evaluation data is included in pre-training corpora used to train large scale models, compromising evaluation results. The workshop fostered a shared task to collect evidence on data contamination in current available datasets and models. The goal of the shared task and associated database is to assist the community in understanding the extent of the problem and to assist researchers in avoiding reporting evaluation results on known contaminated resources. The shared task provides a structured, centralized public database for the collection of contamination evidence, open to contributions from the community via GitHub pool requests. This first compilation paper is based on 566 reported entries over 91 contaminated sources from a total of 23 contributors. The details of the individual contamination events are available in the platform. The platform continues to be online, open to contributions from the community.

Alon Jacovi

The XAI literature is decentralized, both in terminology and in publication venues, but recent years saw the community converge around keywords that make it possible to more reliably discover papers automatically. We use keyword search using the SemanticScholar API and manual curation to collect a well-formatted and reasonably comprehensive set of 5199 XAI papers, available at https://github.com/alonjacovi/XAI-Scholar . We use this collection to clarify and visualize trends about the size and scope of the literature, citation trends, cross-field trends, and collaboration trends. Overall, XAI is becoming increasingly multidisciplinary, with relative growth in papers belonging to increasingly diverse (non-CS) scientific fields, increasing cross-field collaborative authorship, increasing cross-field citation activity. The collection can additionally be used as a paper discovery engine, by retrieving XAI literature which is cited according to specific constraints (for example, papers that are influential outside of their field, or influential to non-XAI research).

Aileen Cheng, Alon Jacovi, Amir Globerson et al.

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models' world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model's overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts .

Tita A. Bach, Jenny K. Kristiansen, Aleksandar Babic et al.

Ensuring quality human-AI interaction (HAII) in safety-critical industries is essential. Failure to do so can lead to catastrophic and deadly consequences. Despite this urgency, existing research on HAII is limited, fragmented, and inconsistent. We present here a survey of that literature and recommendations for research best practices that should improve the field. We divided our investigation into the following areas: 1) terms used to describe HAII, 2) primary roles of AI-enabled systems, 3) factors that influence HAII, and 4) how HAII is measured. Additionally, we described the capabilities and maturity of the AI-enabled systems used in safety-critical industries discussed in these articles. We found that no single term is used across the literature to describe HAII and some terms have multiple meanings. According to our literature, seven factors influence HAII: user characteristics (e.g., user personality), user perceptions and attitudes (e.g., user biases), user expectations and experience (e.g., mismatched user expectations and experience), AI interface and features (e.g., interactive design), AI output (e.g., perceived accuracy), explainability and interpretability (e.g., level of detail, user understanding), and usage of AI (e.g., heterogeneity of environments). HAII is most measured with user-related subjective metrics (e.g., user perceptions, trust, and attitudes), and AI-assisted decision-making is the most common primary role of AI-enabled systems. Based on this review, we conclude that there are substantial research gaps in HAII. Researchers and developers need to codify HAII terminology, involve users throughout the AI lifecycle (especially during development), and tailor HAII in safety-critical industries to the users and environments.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Antonin Poché, Alon Jacovi, Agustin Martin Picard et al.

Concept-based explanations work by mapping complex model computations to human-understandable concepts. Evaluating such explanations is very difficult, as it includes not only the quality of the induced space of possible concepts but also how effectively the chosen concepts are communicated to users. Existing evaluation metrics often focus solely on the former, neglecting the latter. We introduce an evaluation framework for measuring concept explanations via automated simulatability: a simulator's ability to predict the explained model's outputs based on the provided explanations. This approach accounts for both the concept space and its interpretation in an end-to-end evaluation. Human studies for simulatability are notoriously difficult to enact, particularly at the scale of a wide, comprehensive empirical evaluation (which is the subject of this work). We propose using large language models (LLMs) as simulators to approximate the evaluation and report various analyses to make such approximations reliable. Our method allows for scalable and consistent evaluation across various models and datasets. We report a comprehensive empirical evaluation using this framework and show that LLMs provide consistent rankings of explanation methods. Code available at https://github.com/AnonymousConSim/ConSim.

Alon Jacovi, Oren Sar Shalom, Yoav Goldberg

We present an analysis into the inner workings of Convolutional Neural Networks (CNNs) for processing text. CNNs used for computer vision can be interpreted by projecting filters into image space, but for discrete sequence inputs CNNs remain a mystery. We aim to understand the method by which the networks process and classify text. We examine common hypotheses to this problem: that filters, accompanied by global max-pooling, serve as ngram detectors. We show that filters may capture several different semantic classes of ngrams by using different activation patterns, and that global max-pooling induces behavior which separates important ngrams from the rest. Finally, we show practical use cases derived from our findings in the form of model interpretability (explaining a trained model by deriving a concrete identity for each filter, bridging the gap between visualization tools in vision tasks and NLP) and prediction interpretability (explaining predictions). Code implementation is available online at github.com/sayaendo/interpreting-cnn-for-text.

Alon Jacovi, Yonatan Bitton, Bernd Bohnet et al. · deepmind

Prompting language models to provide step-by-step answers (e.g., "Chain-of-Thought") is the prominent approach for complex reasoning tasks, where more accurate reasoning chains typically improve downstream task performance. Recent literature discusses automatic methods to verify reasoning to evaluate and improve their correctness. However, no fine-grained step-level datasets are available to enable thorough evaluation of such verification methods, hindering progress in this direction. We introduce REVEAL: Reasoning Verification Evaluation, a dataset to benchmark automatic verifiers of complex Chain-of-Thought reasoning in open-domain question-answering settings. REVEAL includes comprehensive labels for the relevance, attribution to evidence passages, and logical correctness of each reasoning step in a language model's answer, across a variety of datasets and state-of-the-art language models. Evaluation on REVEAL shows that verifiers struggle at verifying reasoning chains - in particular, verifying logical correctness and detecting contradictions. Available at https://reveal-dataset.github.io/ .

Alon Jacovi, Andrew Wang, Chris Alberti et al. · deepmind

We introduce FACTS Grounding, an online leaderboard and associated benchmark that evaluates language models' ability to generate text that is factually accurate with respect to given context in the user prompt. In our benchmark, each prompt includes a user request and a full document, with a maximum length of 32k tokens, requiring long-form responses. The long-form responses are required to be fully grounded in the provided context document while fulfilling the user request. Models are evaluated using automated judge models in two phases: (1) responses are disqualified if they do not fulfill the user request; (2) they are judged as accurate if the response is fully grounded in the provided document. The automated judge models were comprehensively evaluated against a held-out test-set to pick the best prompt template, and the final factuality score is an aggregate of multiple judge models to mitigate evaluation bias. The FACTS Grounding leaderboard will be actively maintained over time, and contains both public and private splits to allow for external participation while guarding the integrity of the leaderboard. It can be found at https://www.kaggle.com/facts-leaderboard.

Arie Cattan, Alon Jacovi, Ori Ram et al.

Retrieval Augmented Generation (RAG) is a commonly used approach for enhancing large language models (LLMs) with relevant and up-to-date information. However, the retrieved sources can often contain conflicting information and it remains unclear how models should address such discrepancies. In this work, we first propose a novel taxonomy of knowledge conflict types in RAG, along with the desired model behavior for each type. We then introduce CONFLICTS, a high-quality benchmark with expert annotations of conflict types in a realistic RAG setting. CONFLICTS is the first benchmark that enables tracking progress on how models address a wide range of knowledge conflicts. We conduct extensive experiments on this benchmark, showing that LLMs often struggle to appropriately resolve conflicts between sources. While prompting LLMs to explicitly reason about the potential conflict in the retrieved documents significantly improves the quality and appropriateness of their responses, substantial room for improvement in future research remains.

Omer Goldman, Alon Jacovi, Aviv Slobodkin et al.

Improvements in language models' capabilities have pushed their applications towards longer contexts, making long-context evaluation and development an active research area. However, many disparate use-cases are grouped together under the umbrella term of "long-context", defined simply by the total length of the model's input, including - for example - Needle-in-a-Haystack tasks, book summarization, and information aggregation. Given their varied difficulty, in this position paper we argue that conflating different tasks by their context length is unproductive. As a community, we require a more precise vocabulary to understand what makes long-context tasks similar or different. We propose to unpack the taxonomy of long-context based on the properties that make them more difficult with longer contexts. We propose two orthogonal axes of difficulty: (I) Diffusion: How hard is it to find the necessary information in the context? (II) Scope: How much necessary information is there to find? We survey the literature on long-context, provide justification for this taxonomy as an informative descriptor, and situate the literature with respect to it. We conclude that the most difficult and interesting settings, whose necessary information is very long and highly diffused within the input, is severely under-explored. By using a descriptive vocabulary and discussing the relevant properties of difficulty in long-context, we can implement more informed research in this area. We call for a careful design of tasks and benchmarks with distinctly long context, taking into account the characteristics that make it qualitatively different from shorter context.

Arie Cattan, Alon Jacovi, Alex Fabrikant et al.

Despite recent advancements in Large Language Models (LLMs), their performance on tasks involving long contexts remains sub-optimal. In this work, we propose DoubleDipper, a novel In-Context-Learning method that automatically generates few-shot examples for long context QA tasks by recycling contexts. Specifically, given a long input context (1-3k tokens) and a query, we generate additional query-output pairs from the given context as few-shot examples, while introducing the context only once. This ensures that the demonstrations are leveraging the same context as the target query while only adding a small number of tokens to the prompt. We further enhance each demonstration by instructing the model to explicitly identify the relevant paragraphs before the answer, which improves performance while providing fine-grained attribution to the answer source. We apply our method on multiple LLMs and obtain substantial improvements (+16 absolute points on average across models) on various QA datasets with long context. Surprisingly, despite introducing only single-hop ICL examples, LLMs successfully generalize to multi-hop long-context QA using our approach.

Avi Caciularu, Alon Jacovi, Eyal Ben-David et al.

Large Language Models (LLMs) often do not perform well on queries that require the aggregation of information across texts. To better evaluate this setting and facilitate modeling efforts, we introduce TACT - Text And Calculations through Tables, a dataset crafted to evaluate LLMs' reasoning and computational abilities using complex instructions. TACT contains challenging instructions that demand stitching information scattered across one or more texts, and performing complex integration on this information to generate the answer. We construct this dataset by leveraging an existing dataset of texts and their associated tables. For each such tables, we formulate new queries, and gather their respective answers. We demonstrate that all contemporary LLMs perform poorly on this dataset, achieving an accuracy below 38%. To pinpoint the difficulties and thoroughly dissect the problem, we analyze model performance across three components: table-generation, Pandas command-generation, and execution. Unexpectedly, we discover that each component presents substantial challenges for current LLMs. These insights lead us to propose a focused modeling framework, which we refer to as IE as a tool. Specifically, we propose to add "tools" for each of the above steps, and implement each such tool with few-shot prompting. This approach shows an improvement over existing prompting techniques, offering a promising direction for enhancing model capabilities in these tasks.

Alon Jacovi, Avi Caciularu, Omer Goldman et al.

Data contamination has become prevalent and challenging with the rise of models pretrained on large automatically-crawled corpora. For closed models, the training data becomes a trade secret, and even for open models, it is not trivial to detect contamination. Strategies such as leaderboards with hidden answers, or using test data which is guaranteed to be unseen, are expensive and become fragile with time. Assuming that all relevant actors value clean test data and will cooperate to mitigate data contamination, what can be done? We propose three strategies that can make a difference: (1) Test data made public should be encrypted with a public key and licensed to disallow derivative distribution; (2) demand training exclusion controls from closed API holders, and protect your test data by refusing to evaluate without them; (3) avoid data which appears with its solution on the internet, and release the web-page context of internet-derived data along with the data. These strategies are practical and can be effective in preventing data contamination.

Alon Jacovi, Hendrik Schuff, Heike Adel et al.

Word-level saliency explanations ("heat maps over words") are often used to communicate feature-attribution in text-based models. Recent studies found that superficial factors such as word length can distort human interpretation of the communicated saliency scores. We conduct a user study to investigate how the marking of a word's neighboring words affect the explainee's perception of the word's importance in the context of a saliency explanation. We find that neighboring words have significant effects on the word's importance rating. Concretely, we identify that the influence changes based on neighboring direction (left vs. right) and a-priori linguistic and computational measures of phrases and collocations (vs. unrelated neighboring words). Our results question whether text-based saliency explanations should be continued to be communicated at word level, and inform future research on alternative saliency explanation methods.

Hendrik Schuff, Alon Jacovi, Heike Adel et al.

While a lot of research in explainable AI focuses on producing effective explanations, less work is devoted to the question of how people understand and interpret the explanation. In this work, we focus on this question through a study of saliency-based explanations over textual data. Feature-attribution explanations of text models aim to communicate which parts of the input text were more influential than others towards the model decision. Many current explanation methods, such as gradient-based or Shapley value-based methods, provide measures of importance which are well-understood mathematically. But how does a person receiving the explanation (the explainee) comprehend it? And does their understanding match what the explanation attempted to communicate? We empirically investigate the effect of various factors of the input, the feature-attribution explanation, and visualization procedure, on laypeople's interpretation of the explanation. We query crowdworkers for their interpretation on tasks in English and German, and fit a GAMM model to their responses considering the factors of interest. We find that people often mis-interpret the explanations: superficial and unrelated factors, such as word length, influence the explainees' importance assignment despite the explanation communicating importance directly. We then show that some of this distortion can be attenuated: we propose a method to adjust saliencies based on model estimates of over- and under-perception, and explore bar charts as an alternative to heatmap saliency visualization. We find that both approaches can attenuate the distorting effect of specific factors, leading to better-calibrated understanding of the explanation.

Alon Jacovi, Jasmijn Bastings, Sebastian Gehrmann et al.

We investigate a formalism for the conditions of a successful explanation of AI. We consider "success" to depend not only on what information the explanation contains, but also on what information the human explainee understands from it. Theory of mind literature discusses the folk concepts that humans use to understand and generalize behavior. We posit that folk concepts of behavior provide us with a "language" that humans understand behavior with. We use these folk concepts as a framework of social attribution by the human explainee - the information constructs that humans are likely to comprehend from explanations - by introducing a blueprint for an explanatory narrative (Figure 1) that explains AI behavior with these constructs. We then demonstrate that many XAI methods today can be mapped to folk concepts of behavior in a qualitative evaluation. This allows us to uncover their failure modes that prevent current methods from explaining successfully - i.e., the information constructs that are missing for any given XAI method, and whose inclusion can decrease the likelihood of misunderstanding AI behavior.

Alon Jacovi, Swabha Swayamdipta, Shauli Ravfogel et al.

Contrastive explanations clarify why an event occurred in contrast to another. They are more inherently intuitive to humans to both produce and comprehend. We propose a methodology to produce contrastive explanations for classification models by modifying the representation to disregard non-contrastive information, and modifying model behavior to only be based on contrastive reasoning. Our method is based on projecting model representation to a latent space that captures only the features that are useful (to the model) to differentiate two potential decisions. We demonstrate the value of contrastive explanations by analyzing two different scenarios, using both high-level abstract concept attribution and low-level input token/span attribution, on two widely used text classification tasks. Specifically, we produce explanations for answering: for which label, and against which alternative label, is some aspect of the input useful? And which aspects of the input are useful for and against particular decisions? Overall, our findings shed light on the ability of label-contrastive explanations to provide a more accurate and finer-grained interpretability of a model's decision.

Alon Jacovi, Ana Marasović, Tim Miller et al.

Trust is a central component of the interaction between people and AI, in that 'incorrect' levels of trust may cause misuse, abuse or disuse of the technology. But what, precisely, is the nature of trust in AI? What are the prerequisites and goals of the cognitive mechanism of trust, and how can we promote them, or assess whether they are being satisfied in a given interaction? This work aims to answer these questions. We discuss a model of trust inspired by, but not identical to, sociology's interpersonal trust (i.e., trust between people). This model rests on two key properties of the vulnerability of the user and the ability to anticipate the impact of the AI model's decisions. We incorporate a formalization of 'contractual trust', such that trust between a user and an AI is trust that some implicit or explicit contract will hold, and a formalization of 'trustworthiness' (which detaches from the notion of trustworthiness in sociology), and with it concepts of 'warranted' and 'unwarranted' trust. We then present the possible causes of warranted trust as intrinsic reasoning and extrinsic behavior, and discuss how to design trustworthy AI, how to evaluate whether trust has manifested, and whether it is warranted. Finally, we elucidate the connection between trust and XAI using our formalization.

Shachar Rosenman, Alon Jacovi, Yoav Goldberg

The process of collecting and annotating training data may introduce distribution artifacts which may limit the ability of models to learn correct generalization behavior. We identify failure modes of SOTA relation extraction (RE) models trained on TACRED, which we attribute to limitations in the data annotation process. We collect and annotate a challenge-set we call Challenging RE (CRE), based on naturally occurring corpus examples, to benchmark this behavior. Our experiments with four state-of-the-art RE models show that they have indeed adopted shallow heuristics that do not generalize to the challenge-set data. Further, we find that alternative question answering modeling performs significantly better than the SOTA models on the challenge-set, despite worse overall TACRED performance. By adding some of the challenge data as training examples, the performance of the model improves. Finally, we provide concrete suggestion on how to improve RE data collection to alleviate this behavior.

Alon Jacovi, Yoav Goldberg

We find that the requirement of model interpretations to be faithful is vague and incomplete. With interpretation by textual highlights as a case-study, we present several failure cases. Borrowing concepts from social science, we identify that the problem is a misalignment between the causal chain of decisions (causal attribution) and the attribution of human behavior to the interpretation (social attribution). We re-formulate faithfulness as an accurate attribution of causality to the model, and introduce the concept of aligned faithfulness: faithful causal chains that are aligned with their expected social behavior. The two steps of causal attribution and social attribution together complete the process of explaining behavior. With this formalization, we characterize various failures of misaligned faithful highlight interpretations, and propose an alternative causal chain to remedy the issues. Finally, we implement highlight explanations of the proposed causal format using contrastive explanations.

Yanai Elazar, Shauli Ravfogel, Alon Jacovi et al.

A growing body of work makes use of probing to investigate the working of neural models, often considered black boxes. Recently, an ongoing debate emerged surrounding the limitations of the probing paradigm. In this work, we point out the inability to infer behavioral conclusions from probing results and offer an alternative method that focuses on how the information is being used, rather than on what information is encoded. Our method, Amnesic Probing, follows the intuition that the utility of a property for a given task can be assessed by measuring the influence of a causal intervention that removes it from the representation. Equipped with this new analysis tool, we can ask questions that were not possible before, e.g. is part-of-speech information important for word prediction? We perform a series of analyses on BERT to answer these types of questions. Our findings demonstrate that conventional probing performance is not correlated to task importance, and we call for increased scrutiny of claims that draw behavioral or causal conclusions from probing results.

Alon Jacovi, Yoav Goldberg

With the growing popularity of deep-learning based NLP models, comes a need for interpretable systems. But what is interpretability, and what constitutes a high-quality interpretation? In this opinion piece we reflect on the current state of interpretability evaluation research. We call for more clearly differentiating between different desired criteria an interpretation should satisfy, and focus on the faithfulness criteria. We survey the literature with respect to faithfulness evaluation, and arrange the current approaches around three assumptions, providing an explicit form to how faithfulness is "defined" by the community. We provide concrete guidelines on how evaluation of interpretation methods should and should not be conducted. Finally, we claim that the current binary definition for faithfulness sets a potentially unrealistic bar for being considered faithful. We call for discarding the binary notion of faithfulness in favor of a more graded one, which we believe will be of greater practical utility.

Alon Jacovi, Gang Niu, Yoav Goldberg et al.

We consider the situation in which a user has collected a small set of documents on a cohesive topic, and they want to retrieve additional documents on this topic from a large collection. Information Retrieval (IR) solutions treat the document set as a query, and look for similar documents in the collection. We propose to extend the IR approach by treating the problem as an instance of positive-unlabeled (PU) learning -- i.e., learning binary classifiers from only positive and unlabeled data, where the positive data corresponds to the query documents, and the unlabeled data is the results returned by the IR engine. Utilizing PU learning for text with big neural networks is a largely unexplored field. We discuss various challenges in applying PU learning to the setting, including an unknown class prior, extremely imbalanced data and large-scale accurate evaluation of models, and we propose solutions and empirically validate them. We demonstrate the effectiveness of the method using a series of experiments of retrieving PubMed abstracts adhering to fine-grained topics. We demonstrate improvements over the base IR solution and other baselines.

Alon Jacovi, Guy Hadash, Einat Kermany et al.

Deep neural networks work well at approximating complicated functions when provided with data and trained by gradient descent methods. At the same time, there is a vast amount of existing functions that programmatically solve different tasks in a precise manner eliminating the need for training. In many cases, it is possible to decompose a task to a series of functions, of which for some we may prefer to use a neural network to learn the functionality, while for others the preferred method would be to use existing black-box functions. We propose a method for end-to-end training of a base neural network that integrates calls to existing black-box functions. We do so by approximating the black-box functionality with a differentiable neural network in a way that drives the base network to comply with the black-box function interface during the end-to-end optimization process. At inference time, we replace the differentiable estimator with its external black-box non-differentiable counterpart such that the base network output matches the input arguments of the black-box function. Using this "Estimate and Replace" paradigm, we train a neural network, end to end, to compute the input to black-box functionality while eliminating the need for intermediate labels. We show that by leveraging the existing precise black-box function during inference, the integrated model generalizes better than a fully differentiable model, and learns more efficiently compared to RL-based methods.

Guy Hadash, Einat Kermany, Boaz Carmeli et al.

Existing applications include a huge amount of knowledge that is out of reach for deep neural networks. This paper presents a novel approach for integrating calls to existing applications into deep learning architectures. Using this approach, we estimate each application's functionality with an estimator, which is implemented as a deep neural network (DNN). The estimator is then embedded into a base network that we direct into complying with the application's interface during an end-to-end optimization process. At inference time, we replace each estimator with its existing application counterpart and let the base network solve the task by interacting with the existing application. Using this 'Estimate and Replace' method, we were able to train a DNN end-to-end with less data and outperformed a matching DNN that did not interact with the external application.