Widely used alignment techniques, such as reinforcement learning from human feedback (RLHF), rely on the ability of humans to supervise model behavior—for example, to evaluate whether a model faithfully followed instructions or generated safe outputs. However, future superhuman models will behave in complex ways too difficult for humans to reliably evaluate; humans will only be able to weakly supervise superhuman models. We study an analogy to this problem: can weak model supervision elicit the full capabilities of a much stronger model? We test this using a range of pretrained language models in the GPT-4 family on natural language processing (NLP), chess, and reward modeling tasks. We find that when we naively finetune strong pretrained models on labels generated by a weak model, they consistently perform better than their weak supervisors, a phenomenon we call weak-to-strong generalization. However, we are still far from recovering the full capabilities of strong models with naive finetuning alone, suggesting that techniques like RLHF may scale poorly to superhuman models without further work. We find that simple methods can often significantly improve weak-to-strong generalization: for example, when finetuning GPT-4 with a GPT-2-level supervisor and an auxiliary confidence loss, we can recover close to GPT-3.5-level performance on NLP tasks. Our results suggest that it is feasible to make empirical progress today on a fundamental challenge of aligning superhuman models.

Recommendation problems with large numbers of discrete items, such as products, webpages, or videos, are ubiquitous in the technology industry. Deep neural networks are being increasingly used for these recommendation problems. These models use embeddings to represent discrete items as continuous vectors, and the vocabulary sizes and embedding dimensions, despite their heavy influence on the model's accuracy, are often manually selected in a heuristical manner.

We presentsmart drill-down, an operator for interactively exploring a relational table to discover and summarize “interesting” groups of tuples. Each group of tuples is described by arule. For instance, the rule$(a, b, \star, 1000)$tells us that there are 1,000 tuples with value$a$in the first column and$b$in the second column (and any value in the third column). Smart drill-down presents an analyst with a list of rules that together describe interesting aspects of the table. The analyst can tailor the definition of interesting, and can interactively apply smart drill-down on an existing rule to explore that part of the table. We demonstrate that the underlying optimization problems areNP-Hard, and describe an algorithm for finding the approximately optimal list of rules to display when the user uses a smart drill-down, and a dynamic sampling scheme for efficiently interacting with large tables. Finally, we perform experiments on real datasets on our experimental prototype to demonstrate the usefulness of smart drill-down and study the performance of our algorithms.

We study the problem of finding and monitoring fixed-size subgraphs in a continually changing large-scale graph. We present the first approach that (i) performs worst-case optimal computation and communication, (ii) maintains a total memory footprint linear in the number of input edges, and (iii) scales down per-worker computation, communication, and memory requirements linearly as the number of workers increases, even on adversarially skewed inputs. Our approach is based on worst-case optimal join algorithms, recast as a data-parallel dataflow computation. We describe the general algorithm and modifications that make it robust to skewed data, prove theoretical bounds on its resource requirements in the massively parallel computing model, and implement and evaluate it on graphs containing as many as 64 billion edges. The underlying algorithm and ideas generalize from finding and monitoring subgraphs to the more general problem of computing and maintaining relational equi-joins over dynamic relations.

We focus on data fusion, i.e., the problem of unifying conflicting data from data sources into a single representation by estimating the source accuracies. We propose SLiMFast, a framework that expresses data fusion as a statistical learning problem over discriminative probabilistic models, which in many cases correspond to logistic regression. In contrast to previous approaches that use complex generative models, discriminative models make fewer distributional assumptions over data sources and allow us to obtain rigorous theoretical guarantees. Furthermore, we show how SLiMFast enables incorporating domain knowledge into data fusion, yielding accuracy improvements of up to 50% over state-of-the-art baselines. Building upon our theoretical results, we design an optimizer that obviates the need for users to manually select an algorithm for learning SLiMFast's parameters. We validate our optimizer on multiple real-world datasets and show that it can accurately predict the learning algorithm that yields the best data fusion results.

We present smart drill-down, an operator for interactively exploring a relational table to discover and summarize “interesting” groups of tuples. Each group of tuples is described by a rule. For instance, the rule (a, b, *, 1000) tells us that there are a thousand tuples with value a in the first column and b in the second column (and any value in the third column). Smart drill-down presents an analyst with a list of rules that together describe interesting aspects of the table. The analyst can tailor the definition of interesting, and can interactively apply smart drill-down on an existing rule to explore that part of the table. We demonstrate that the underlying optimization problems are NP-HARD, and describe an algorithm for finding the approximately optimal list of rules to display when the user uses a smart drill-down, and a dynamic sampling scheme for efficiently interacting with large tables. Finally, we perform experiments on real datasets on our experimental prototype to demonstrate the usefulness of smart drill-down and study the performance of our algorithms.

Multiround algorithms are now commonly used in distributed data processing systems, yet the extent to which algorithms can benefit from running more rounds is not well understood. This paper answers this question for several rounds for the problem of computing the equijoin of n relations. Given any query Q with width w, intersection width iw, input size IN, output size OUT, and a cluster of machines with M=\Omega(IN \frac{1}{\epsilon}) memory available per machine, where \epsilon > 1 and w \ge 1 are constants, we show that: 1. Q can be computed in O(n) rounds with O(n(INw + OUT)2/M) communication cost with high probability. Q can be computed in O(log(n)) rounds with O(n(INmax(w, 3iw) + OUT)2/M) communication cost with high probability. Intersection width is a new notion we introduce for queries and generalized hypertree decompositions (GHDs) of queries that captures how connected the adjacent components of the GHDs are. We achieve our first result by introducing a distributed and generalized version of Yannakakis's algorithm, called GYM. GYM takes as input any GHD of Q with width w and depth d, and computes Q in O(d + log(n)) rounds and O(n (INw + OUT)2/M) communication cost. We achieve our second result by showing how to construct GHDs of Q with width max(w, 3iw) and depth O(log(n)). We describe another technique to construct GHDs with longer widths and lower depths, demonstrating other tradeoffs one can make between communication and the number of rounds.

We study a class of aggregate-join queries with multiple aggregation operators evaluated over annotated relations. We show that straightforward extensions of standard multiway join algorithms and generalized hypertree decompositions (GHDs) provide best-known runtime guarantees. In contrast, prior work uses bespoke algorithms and data structures and does not match these guarantees. We extend the standard techniques by providing a complete characterization of (1) the set of orderings equivalent to a given ordering and (2) the set of GHDs valid with respect to the given ordering, i.e., GHDs that correctly answer a given aggregate-join query when provided to (simple variants of) standard join algorithms. We show by example that previous approaches are incomplete. The key technical consequence of our characterizations is a decomposition of a valid GHD into a set of (smaller) unconstrained GHDs, i.e., into a set of GHDs of sub-queries without aggregations. Since this decomposition is comprised of unconstrained GHDs, we are able to connect to the wide literature on GHDs for join query processing, thereby obtaining improved runtime bounds, MapReduce variants, and an efficient method to find approximately optimal GHDs.

Crowdsourcing refers to solving large problems by involving human workers that solve component sub-problems or tasks. In data crowdsourcing, the problem involves data acquisition, management, and analysis. In this paper, we provide an overview of data crowdsourcing, giving examples of problems that the authors have tackled, and presenting the key design steps involved in implementing a crowdsourced solution. We also discuss some of the open challenges that remain to be solved.

Cipherbase is a comprehensive database system that provides strong end-to-end data confidentiality through encryption. Cipherbase is based on a novel architecture that combines an industrial strength database engine (SQL Server) with lightweight processing over encrypted data that is performed in secure hardware. The overall architecture provides significant benefits over the state-of-the-art in terms of security, performance, and functionality. This paper presents a prototype of Cipherbase that uses FPGAs to provide secure processing and describes the system engineering details implemented to achieve competitive performance for transactional workloads. This includes hardware-software co-design issues (e.g. how to best offer parallelism), optimizations to hide the latency between the secure hardware and the main system, and techniques to cope with space inefficiencies. All these optimizations were carefully designed not to affect end-to-end data confidentiality. Our experiments with the TPC-C benchmark show that in the worst case when all data are strongly encrypted, Cipherbase achieves 40% of the throughput of plaintext SQL Server. In more realistic cases, if only critical data such as customer names are encrypted, the Cipherbase throughput is more than 90% of plaintext SQL Server.

Evaluating workers is a critical aspect of any crowdsourcing system. In this paper, we devise techniques for evaluating workers by finding confidence intervals on their error rates. Unlike prior work, we focus on “conciseness”-that is, giving as tight a confidence interval as possible. Conciseness is of utmost importance because it allows us to be sure that we have the best guarantee possible on worker error rate. Also unlike prior work, we provide techniques that work under very general scenarios, such as when not all workers have attempted every task (a fairly common scenario in practice), when tasks have non-boolean responses, and when workers have different biases for positive and negative tasks. We demonstrate conciseness as well as accuracy of our confidence intervals by testing them on a variety of conditions and multiple real-world datasets.

User Defined Function(UDFs) are used increasingly to augment query languages with extra, application dependent functionality. Selection queries involving UDF predicates tend to be expensive, either in terms of monetary cost or latency. In this paper, we study ways to efficiently evaluate selection queries with UDF predicates. We provide a family of techniques for processing queries at low cost while satisfying user-specified precision and recall constraints. Our techniques are applicable to a variety of scenarios including when selection probabilities of tuples are available beforehand, when this information is available but noisy, or when no such prior information is available. We also generalize our techniques to more complex queries. Finally, we test our techniques on real datasets, and show that they achieve significant savings in UDF evaluations of up to $80\%$, while incurring only a small reduction in accuracy.

We present a data exploration system equipped with smart drill-down , a novel operator for interactively exploring a relational table to discover and summarize "interesting" groups of tuples. Each such group of tuples is represented by a rule. For instance, the rule ( a, b , *, 1000) tells us that there are a thousand tuples with value a in the first column and b in the second column (and any value in the third column). Smart drill-down presents an analyst with a list of rules that together describe interesting aspects of the table. The analyst can tailor the definition of interesting, and can interactively apply smart drill-down on an existing rule to explore that part of the table. In the demonstration, conference attendees will be able to use the data exploration system equipped with smart drill-down, and will be able to contrast smart drill-down to traditional drill-down, for various interestingness measures, and resource constraints.

Worker quality control is a crucial aspect of crowdsourcing systems; typically occupying a large fraction of the time and money invested on crowdsourcing. In this work, we devise techniques to generate confidence intervals for worker error rate estimates, thereby enabling a better evaluation of worker quality. We show that our techniques generate correct confidence intervals on a range of real-world datasets, and demonstrate wide applicability by using them to evict poorly performing workers, and provide confidence intervals on the accuracy of the answers.

Data confidentiality is one of the main concerns for users of public cloud services. The key problem is protecting sensitive data from being accessed by cloud administrators who have root privileges and can remotely inspect the memory and disk contents of the cloud servers. While encryption is the basic mechanism that can leveraged to provide data confidentiality, providing an efficient database-as-a-service that can run on encrypted data raises several interesting challenges. In this demonstration we outline the functionality of Cipherbase --- a full fledged SQL database system that supports the full generality of a database system while providing high data confidentiality. Cipherbase has a novel architecture that tightly integrates custom-designed trusted hardware for performing operations on encrypted data securely such that an administrator cannot get access to any plaintext corresponding to sensitive data.

We consider Markov decision processes (MDPs) with specifications given as Büchi (liveness) objectives. We consider the problem of computing the set of almost-sure winning vertices from where the objective can be ensured with probability 1. We study for the first time the average case complexity of the classical algorithm for computing the set of almost-sure winning vertices for MDPs with Buchi objectives. Our contributions are as follows: First, we show that for MDPs with constant out-degree the expected number of iterations is at most logarithmic and the average case running time is linear (as compared to the worst case linear number of iterations and quadratic time complexity). Second, for the average case analysis over all MDPs we show that the expected number of iterations is constant and the average case running time is linear (again as compared to the worst case linear number of iterations and quadratic time complexity). Finally we also show that given that all MDPs are equally likely, the probability that the classical algorithm requires more than constant number of iterations is exponentially small.