Eran Yahav

Shaked Brody, Uri Alon, Eran Yahav

Layer Normalization (LayerNorm) is an inherent component in all Transformer-based models. In this paper, we show that LayerNorm is crucial to the expressivity of the multi-head attention layer that follows it. This is in contrast to the common belief that LayerNorm's only role is to normalize the activations during the forward pass, and their gradients during the backward pass. We consider a geometric interpretation of LayerNorm and show that it consists of two components: (a) projection of the input vectors to a $d-1$ space that is orthogonal to the $\left[1,1,...,1\right]$ vector, and (b) scaling of all vectors to the same norm of $\sqrt{d}$. We show that each of these components is important for the attention layer that follows it in Transformers: (a) projection allows the attention mechanism to create an attention query that attends to all keys equally, offloading the need to learn this operation by the attention; and (b) scaling allows each key to potentially receive the highest attention, and prevents keys from being "un-select-able". We show empirically that Transformers do indeed benefit from these properties of LayeNorm in general language modeling and even in computing simple functions such as "majority". Our code is available at https://github.com/tech-srl/layer_norm_expressivity_role .

Shaked Brody, Uri Alon, Eran Yahav

Graph Attention Networks (GATs) are one of the most popular GNN architectures and are considered as the state-of-the-art architecture for representation learning with graphs. In GAT, every node attends to its neighbors given its own representation as the query. However, in this paper we show that GAT computes a very limited kind of attention: the ranking of the attention scores is unconditioned on the query node. We formally define this restricted kind of attention as static attention and distinguish it from a strictly more expressive dynamic attention. Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. To remove this limitation, we introduce a simple fix by modifying the order of operations and propose GATv2: a dynamic graph attention variant that is strictly more expressive than GAT. We perform an extensive evaluation and show that GATv2 outperforms GAT across 11 OGB and other benchmarks while we match their parametric costs. Our code is available at https://github.com/tech-srl/how_attentive_are_gats . GATv2 is available as part of the PyTorch Geometric library, the Deep Graph Library, and the TensorFlow GNN library.

Uri Alon, Eran Yahav

Since the proposal of the graph neural network (GNN) by Gori et al. (2005) and Scarselli et al. (2008), one of the major problems in training GNNs was their struggle to propagate information between distant nodes in the graph. We propose a new explanation for this problem: GNNs are susceptible to a bottleneck when aggregating messages across a long path. This bottleneck causes the over-squashing of exponentially growing information into fixed-size vectors. As a result, GNNs fail to propagate messages originating from distant nodes and perform poorly when the prediction task depends on long-range interaction. In this paper, we highlight the inherent problem of over-squashing in GNNs: we demonstrate that the bottleneck hinders popular GNNs from fitting long-range signals in the training data; we further show that GNNs that absorb incoming edges equally, such as GCN and GIN, are more susceptible to over-squashing than GAT and GGNN; finally, we show that prior work, which extensively tuned GNN models of long-range problems, suffers from over-squashing, and that breaking the bottleneck improves their state-of-the-art results without any tuning or additional weights. Our code is available at https://github.com/tech-srl/bottleneck/ .

Shaked Brody, Uri Alon, Eran Yahav

We address the problem of predicting edit completions based on a learned model that was trained on past edits. Given a code snippet that is partially edited, our goal is to predict a completion of the edit for the rest of the snippet. We refer to this task as the EditCompletion task and present a novel approach for tackling it. The main idea is to directly represent structural edits. This allows us to model the likelihood of the edit itself, rather than learning the likelihood of the edited code. We represent an edit operation as a path in the program's Abstract Syntax Tree (AST), originating from the source of the edit to the target of the edit. Using this representation, we present a powerful and lightweight neural model for the EditCompletion task. We conduct a thorough evaluation, comparing our approach to a variety of representation and modeling approaches that are driven by multiple strong models such as LSTMs, Transformers, and neural CRFs. Our experiments show that our model achieves a 28% relative gain over state-of-the-art sequential models and 2x higher accuracy than syntactic models that learn to generate the edited code, as opposed to modeling the edits directly. Our code, dataset, and trained models are publicly available at https://github.com/tech-srl/c3po/ .

Noam Yefet, Uri Alon, Eran Yahav

Neural models of code have shown impressive results when performing tasks such as predicting method names and identifying certain kinds of bugs. We show that these models are vulnerable to adversarial examples, and introduce a novel approach for attacking trained models of code using adversarial examples. The main idea of our approach is to force a given trained model to make an incorrect prediction, as specified by the adversary, by introducing small perturbations that do not change the program's semantics, thereby creating an adversarial example. To find such perturbations, we present a new technique for Discrete Adversarial Manipulation of Programs (DAMP). DAMP works by deriving the desired prediction with respect to the model's inputs, while holding the model weights constant, and following the gradients to slightly modify the input code. We show that our DAMP attack is effective across three neural architectures: code2vec, GGNN, and GNN-FiLM, in both Java and C#. Our evaluations demonstrate that DAMP has up to 89% success rate in changing a prediction to the adversary's choice (a targeted attack) and a success rate of up to 94% in changing a given prediction to any incorrect prediction (a non-targeted attack). To defend a model against such attacks, we empirically examine a variety of possible defenses and discuss their trade-offs. We show that some of these defenses can dramatically drop the success rate of the attacker, with a minor penalty of 2% relative degradation in accuracy when they are not performing under attack. Our code, data, and trained models are available at https://github.com/tech-srl/adversarial-examples .

Uri Alon, Roy Sadaka, Omer Levy et al.

We address the problem of any-code completion - generating a missing piece of source code in a given program without any restriction on the vocabulary or structure. We introduce a new approach to any-code completion that leverages the strict syntax of programming languages to model a code snippet as a tree - structural language modeling (SLM). SLM estimates the probability of the program's abstract syntax tree (AST) by decomposing it into a product of conditional probabilities over its nodes. We present a neural model that computes these conditional probabilities by considering all AST paths leading to a target node. Unlike previous techniques that have severely restricted the kinds of expressions that can be generated in this task, our approach can generate arbitrary code in any programming language. Our model significantly outperforms both seq2seq and a variety of structured approaches in generating Java and C# code. Our code, data, and trained models are available at http://github.com/tech-srl/slm-code-generation/ . An online demo is available at http://AnyCodeGen.org .

Yaniv David, Uri Alon, Eran Yahav

We address the problem of reverse engineering of stripped executables, which contain no debug information. This is a challenging problem because of the low amount of syntactic information available in stripped executables, and the diverse assembly code patterns arising from compiler optimizations. We present a novel approach for predicting procedure names in stripped executables. Our approach combines static analysis with neural models. The main idea is to use static analysis to obtain augmented representations of call sites; encode the structure of these call sites using the control-flow graph (CFG) and finally, generate a target name while attending to these call sites. We use our representation to drive graph-based, LSTM-based and Transformer-based architectures. Our evaluation shows that our models produce predictions that are difficult and time consuming for humans, while improving on existing methods by 28% and by 100% over state-of-the-art neural textual models that do not use any static analysis. Code and data for this evaluation are available at https://github.com/tech-srl/Nero .

Uri Alon, Shaked Brody, Omer Levy et al.

The ability to generate natural language sequences from source code snippets has a variety of applications such as code summarization, documentation, and retrieval. Sequence-to-sequence (seq2seq) models, adopted from neural machine translation (NMT), have achieved state-of-the-art performance on these tasks by treating source code as a sequence of tokens. We present ${\rm {\scriptsize CODE2SEQ}}$: an alternative approach that leverages the syntactic structure of programming languages to better encode source code. Our model represents a code snippet as the set of compositional paths in its abstract syntax tree (AST) and uses attention to select the relevant paths while decoding. We demonstrate the effectiveness of our approach for two tasks, two programming languages, and four datasets of up to $16$M examples. Our model significantly outperforms previous models that were specifically designed for programming languages, as well as state-of-the-art NMT models. An interactive online demo of our model is available at http://code2seq.org. Our code, data and trained models are available at http://github.com/tech-srl/code2seq.

Uri Alon, Meital Zilberstein, Omer Levy et al.

We present a neural model for representing snippets of code as continuous distributed vectors ("code embeddings"). The main idea is to represent a code snippet as a single fixed-length $\textit{code vector}$, which can be used to predict semantic properties of the snippet. This is performed by decomposing code to a collection of paths in its abstract syntax tree, and learning the atomic representation of each path $\textit{simultaneously}$ with learning how to aggregate a set of them. We demonstrate the effectiveness of our approach by using it to predict a method's name from the vector representation of its body. We evaluate our approach by training a model on a dataset of 14M methods. We show that code vectors trained on this dataset can predict method names from files that were completely unobserved during training. Furthermore, we show that our model learns useful method name vectors that capture semantic similarities, combinations, and analogies. Comparing previous techniques over the same data set, our approach obtains a relative improvement of over 75%, being the first to successfully predict method names based on a large, cross-project, corpus. Our trained model, visualizations and vector similarities are available as an interactive online demo at http://code2vec.org. The code, data, and trained models are available at https://github.com/tech-srl/code2vec.

Daniel Glickman, Eran Yahav

The dominant paradigm for machine learning on graphs uses Message Passing Graph Neural Networks (MP-GNNs), in which node representations are updated by aggregating information in their local neighborhood. Recently, there have been increasingly more attempts to adapt the Transformer architecture to graphs in an effort to solve some known limitations of MP-GNN. A challenging aspect of designing Graph Transformers is integrating the arbitrary graph structure into the architecture. We propose Graph Diffuser (GD) to address this challenge. GD learns to extract structural and positional relationships between distant nodes in the graph, which it then uses to direct the Transformer's attention and node representation. We demonstrate that existing GNNs and Graph Transformers struggle to capture long-range interactions and how Graph Diffuser does so while admitting intuitive visualizations. Experiments on eight benchmarks show Graph Diffuser to be a highly competitive model, outperforming the state-of-the-art in a diverse set of domains.

Gail Weiss, Yoav Goldberg, Eran Yahav

What is the computational model behind a Transformer? Where recurrent neural networks have direct parallels in finite state machines, allowing clear discussion and thought around architecture variants or trained models, Transformers have no such familiar parallel. In this paper we aim to change that, proposing a computational model for the transformer-encoder in the form of a programming language. We map the basic components of a transformer-encoder -- attention and feed-forward computation -- into simple primitives, around which we form a programming language: the Restricted Access Sequence Processing Language (RASP). We show how RASP can be used to program solutions to tasks that could conceivably be learned by a Transformer, and how a Transformer can be trained to mimic a RASP solution. In particular, we provide RASP programs for histograms, sorting, and Dyck-languages. We further use our model to relate their difficulty in terms of the number of required layers and attention heads: analyzing a RASP program implies a maximum number of heads and layers necessary to encode a task in a transformer. Finally, we see how insights gained from our abstraction might be used to explain phenomena seen in recent works.

William Merrill, Gail Weiss, Yoav Goldberg et al.

We develop a formal hierarchy of the expressive capacity of RNN architectures. The hierarchy is based on two formal properties: space complexity, which measures the RNN's memory, and rational recurrence, defined as whether the recurrent update can be described by a weighted finite-state machine. We place several RNN variants within this hierarchy. For example, we prove the LSTM is not rational, which formally separates it from the related QRNN (Bradbury et al., 2016). We also show how these models' expressive capacity is expanded by stacking multiple layers or composing them with different pooling functions. Our results build on the theory of "saturated" RNNs (Merrill, 2019). While formally extending these findings to unsaturated RNNs is left to future work, we hypothesize that the practical learnable capacity of unsaturated RNNs obeys a similar hierarchy. Experimental findings from training unsaturated networks on formal languages support this conjecture.

Gail Weiss, Yoav Goldberg, Eran Yahav

We present an algorithm for extraction of a probabilistic deterministic finite automaton (PDFA) from a given black-box language model, such as a recurrent neural network (RNN). The algorithm is a variant of the exact-learning algorithm L*, adapted to a probabilistic setting with noise. The key insight is the use of conditional probabilities for observations, and the introduction of a local tolerance when comparing them. When applied to RNNs, our algorithm often achieves better word error rate (WER) and normalised distributed cumulative gain (NDCG) than that achieved by spectral extraction of weighted finite automata (WFA) from the same networks. PDFAs are substantially more expressive than n-grams, and are guaranteed to be stochastic and deterministic - unlike spectrally extracted WFAs.

Omer Katz, Yuval Olshaker, Yoav Goldberg et al.

We address the problem of automatic decompilation, converting a program in low-level representation back to a higher-level human-readable programming language. The problem of decompilation is extremely important for security researchers. Finding vulnerabilities and understanding how malware operates is much easier when done over source code. The importance of decompilation has motivated the construction of hand-crafted rule-based decompilers. Such decompilers have been designed by experts to detect specific control-flow structures and idioms in low-level code and lift them to source level. The cost of supporting additional languages or new language features in these models is very high. We present a novel approach to decompilation based on neural machine translation. The main idea is to automatically learn a decompiler from a given compiler. Given a compiler from a source language S to a target language T , our approach automatically trains a decompiler that can translate (decompile) T back to S . We used our framework to decompile both LLVM IR and x86 assembly to C code with high success rates. Using our LLVM and x86 instantiations, we were able to successfully decompile over 97% and 88% of our benchmarks respectively.

Gail Weiss, Yoav Goldberg, Eran Yahav

While Recurrent Neural Networks (RNNs) are famously known to be Turing complete, this relies on infinite precision in the states and unbounded computation time. We consider the case of RNNs with finite precision whose computation time is linear in the input length. Under these limitations, we show that different RNN variants have different computational power. In particular, we show that the LSTM and the Elman-RNN with ReLU activation are strictly stronger than the RNN with a squashing activation and the GRU. This is achieved because LSTMs and ReLU-RNNs can easily implement counting behavior. We show empirically that the LSTM does indeed learn to effectively use the counting mechanism.

Uri Alon, Meital Zilberstein, Omer Levy et al.

Predicting program properties such as names or expression types has a wide range of applications. It can ease the task of programming and increase programmer productivity. A major challenge when learning from programs is $\textit{how to represent programs in a way that facilitates effective learning}$. We present a $\textit{general path-based representation}$ for learning from programs. Our representation is purely syntactic and extracted automatically. The main idea is to represent a program using paths in its abstract syntax tree (AST). This allows a learning model to leverage the structured nature of code rather than treating it as a flat sequence of tokens. We show that this representation is general and can: (i) cover different prediction tasks, (ii) drive different learning algorithms (for both generative and discriminative models), and (iii) work across different programming languages. We evaluate our approach on the tasks of predicting variable names, method names, and full types. We use our representation to drive both CRF-based and word2vec-based learning, for programs of four languages: JavaScript, Java, Python and C\#. Our evaluation shows that our approach obtains better results than task-specific handcrafted representations across different tasks and programming languages.

Gail Weiss, Yoav Goldberg, Eran Yahav

We present a novel algorithm that uses exact learning and abstraction to extract a deterministic finite automaton describing the state dynamics of a given trained RNN. We do this using Angluin's L* algorithm as a learner and the trained RNN as an oracle. Our technique efficiently extracts accurate automata from trained RNNs, even when the state vectors are large and require fine differentiation.

Nader H. Bshouty, Dana Drachsler-Cohen, Martin Vechev et al.

Let $F$ be a set of boolean functions. We present an algorithm for learning $F_\vee := \{\vee_{f\in S} f \mid S \subseteq F\}$ from membership queries. Our algorithm asks at most $|F| \cdot OPT(F_\vee)$ membership queries where $OPT(F_\vee)$ is the minimum worst case number of membership queries for learning $F_\vee$. When $F$ is a set of halfspaces over a constant dimension space or a set of variable inequalities, our algorithm runs in polynomial time. The problem we address has practical importance in the field of program synthesis, where the goal is to synthesize a program that meets some requirements. Program synthesis has become popular especially in settings aiming to help end users. In such settings, the requirements are not provided upfront and the synthesizer can only learn them by posing membership queries to the end user. Our work enables such synthesizers to learn the exact requirements while bounding the number of membership queries.

Meital Ben Sinai, Nimrod Partush, Shir Yadid et al.

We present an effective Sybil attack against social location based services. Our attack is based on creating a large number of reputed "bot drivers", and controlling their reported locations using fake GPS reports. We show how this attack can be used to influence social navigation systems by applying it to Waze - a prominent social navigation application used by over 50 million drivers. We show that our attack can fake traffic jams and dramatically influence routing decisions. We present several techniques for preventing the attack, and show that effective mitigation likely requires the use of additional carrier information.