搜索结果：Erases

Delta-rule linear attention improves recurrent memory updates by correcting what is already stored at the current write address before writing new content. However, the active correction is still anchored to that same write address. As a result, stale information stored at a different address cannot be actively removed before new content is written elsewhere. We propose Erase-then-Delta Attention (EDA), a memory update rule that decouples where to erase from where to write. The key insight is that recurrent memory models should not only correct the current write, but also selectively suppress outdated memory at an independently chosen address. Concretely, our method first applies a targeted erase step along a learned erase direction, and then performs the standard delta-style corrective write along the current write direction. This preserves the corrective behavior of delta-rule updates while expanding their memory-management capacity. Language-model pretraining experiments across dense 2.5B and MoE 25B-A2.8B model families show that EDA performs best in both settings. The gain persists after 80B-token long-context midtraining of the MoE models, where EDA also performs best in long

Quantum key distribution protocols based on the quantum eraser phenomenon offer an operational advantage: automatic identification of matching and mismatching encoding choices through interference, eliminating basis reconciliation. However, binary quantum eraser implementations permit an eavesdropper to recover Alice's encoded bit with $85\%$ probability. To overcome this constraint, we introduce a ternary quantum eraser protocol employing three polarization states with $120^\circ$ angular separation, transmitted in three-photon groups with randomized temporal ordering. This extension achieves enhanced security through two complementary mechanisms. First, the reduced distinguishability of symmetrically-arranged quantum states limits single-photon discrimination. Second, the combinatorial complexity of unknown photon ordering constrains multi-photon eavesdropping strategies. Security analysis against individual eavesdropping attacks within the four-dimensional path-polarization Hilbert space establishes that an eavesdropper's maximum success probability is bounded at $54\%$, substantially below the binary discrimination bound. The protocol maintains a binary-equivalent efficiency of

While search-augmented large language models (LLMs) exhibit impressive capabilities, their reliability in complex multi-hop reasoning remains limited. This limitation arises from three fundamental challenges: decomposition errors, where tasks are incorrectly broken down; retrieval missing, where key evidence fails to be retrieved; and reasoning errors, where flawed logic propagates through the reasoning chain. A single failure in any of these stages can derail the final answer. We propose Erasable Reinforcement Learning (ERL), a novel framework that transforms fragile reasoning into a robust process. ERL explicitly identifies faulty steps, erases them, and regenerates reasoning in place, preventing defective logic from propagating through the reasoning chain. This targeted correction mechanism turns brittle reasoning into a more resilient process. Models trained with ERL, termed ESearch, achieve substantial improvements on HotpotQA, MuSiQue, 2Wiki, and Bamboogle, with the 3B model achieving +8.48% EM and +11.56% F1, and the 7B model achieving +5.38% EM and +7.22% F1 over previous state-of-the-art(SOTA) results. These findings suggest that erasable reinforcement learning provides a

Recent advances in text-to-video (T2V) diffusion models have significantly enhanced the quality of generated videos. However, their capability to produce explicit or harmful content introduces new challenges related to misuse and potential rights violations. To address this newly emerging threat, we propose unlearning-based concept erasing as a solution. First, we adopt negatively-guided velocity prediction fine-tuning and enhance it with prompt augmentation to ensure robustness against prompts refined by large language models (LLMs). Second, to achieve precise unlearning, we incorporate mask-based localization regularization and concept preservation regularization to preserve the model's ability to generate non-target concepts. Extensive experiments demonstrate that our method effectively erases a specific concept while preserving the model's generation capability for all other concepts, outperforming existing methods. We provide the unlearned models in \href{https://github.com/VDIGPKU/T2VUnlearning.git}{https://github.com/VDIGPKU/T2VUnlearning.git}.

Large language models trained on web-scale data can memorize private or sensitive knowledge, raising significant privacy risks. Although some unlearning methods mitigate these risks, they remain vulnerable to "relearning" during subsequent training, allowing a substantial portion of forgotten knowledge to resurface. In this paper, we show that widely used unlearning methods cause shallow alignment: instead of faithfully erasing target knowledge, they generate spurious unlearning neurons that amplify negative influence to hide it. To overcome this limitation, we introduce Ssiuu, a new class of unlearning methods that employs attribution-guided regularization to prevent spurious negative influence and faithfully remove target knowledge. Experimental results confirm that our method reliably erases target knowledge and outperforms strong baselines across two practical retraining scenarios: (1) adversarial injection of private data, and (2) benign attack using an instruction-following benchmark. Our findings highlight the necessity of robust and faithful unlearning methods for safe deployment of language models.

The delayed-choice quantum eraser represents an interesting experiment that exemplifies Bohr's principle of complementarity in a beautiful way. According to the complementarity principle, in a two-path interference experiment, the knowledge of which path was taken by the particle and the appearance of interference are mutually exclusive. Even when the which-path information is merely retained in specific quantum path-markers, without being actually read, it suffices to eliminate interference. Nevertheless, if this path information is ``erased'' in some manner, the interference re-emerges, a phenomenon referred to as the quantum eraser. An intriguing aspect of this experiment is that if the path information is erased \emph{after} the particle has been detected on the screen, the interference still reappears, a phenomenon known as the delayed-choice quantum eraser. This observation has led to the interpretation that the particle can be influenced to exhibit characteristics of either a particle or a wave based on a decision made long after it has been registered on the screen. This idea has sparked considerable debate and discussions surrounding retrocausality. This controversy is rev

Federated learning (FL) is emerging as a promising technique for collaborative learning without local data leaving their devices. However, clients' data originating from diverse domains may degrade model performance due to domain shifts, preventing the model from learning consistent representation space. In this paper, we propose a novel FL framework, Federated Domain Shift Eraser (FDSE), to improve model performance by differently erasing each client's domain skew and enhancing their consensus. First, we formulate the model forward passing as an iterative deskewing process that extracts and then deskews features alternatively. This is efficiently achieved by decomposing each original layer in the neural network into a Domain-agnostic Feature Extractor (DFE) and a Domain-specific Skew Eraser (DSE). Then, a regularization term is applied to promise the effectiveness of feature deskewing by pulling local statistics of DSE's outputs close to the globally consistent ones. Finally, DFE modules are fairly aggregated and broadcast to all the clients to maximize their consensus, and DSE modules are personalized for each client via similarity-aware aggregation to erase their domain skew dif

In this work, we present DeepEraser, an effective deep network for generic text removal. DeepEraser utilizes a recurrent architecture that erases the text in an image via iterative operations. Our idea comes from the process of erasing pencil script, where the text area designated for removal is subject to continuous monitoring and the text is attenuated progressively, ensuring a thorough and clean erasure. Technically, at each iteration, an innovative erasing module is deployed, which not only explicitly aggregates the previous erasing progress but also mines additional semantic context to erase the target text. Through iterative refinements, the text regions are progressively replaced with more appropriate content and finally converge to a relatively accurate status. Furthermore, a custom mask generation strategy is introduced to improve the capability of DeepEraser for adaptive text removal, as opposed to indiscriminately removing all the text in an image. Our DeepEraser is notably compact with only 1.4M parameters and trained in an end-to-end manner. To verify its effectiveness, extensive experiments are conducted on several prevalent benchmarks, including SCUT-Syn, SCUT-EnsTex

Despite the impressive capabilities of generating images, text-to-image diffusion models are susceptible to producing undesirable outputs such as NSFW content and copyrighted artworks. To address this issue, recent studies have focused on fine-tuning model parameters to erase problematic concepts. However, existing methods exhibit a major flaw in robustness, as fine-tuned models often reproduce the undesirable outputs when faced with cleverly crafted prompts. This reveals a fundamental limitation in the current approaches and may raise risks for the deployment of diffusion models in the open world. To address this gap, we locate the concept-correlated neurons and find that these neurons show high sensitivity to adversarial prompts, thus could be deactivated when erasing and reactivated again under attacks. To improve the robustness, we introduce a new pruning-based strategy for concept erasing. Our method selectively prunes critical parameters associated with the concepts targeted for removal, thereby reducing the sensitivity of concept-related neurons. Our method can be easily integrated with existing concept-erasing techniques, offering a robust improvement against adversarial in

This work investigates a new erase scheme in NAND flash memory to improve the lifetime and performance of modern solid-state drives (SSDs). In NAND flash memory, an erase operation applies a high voltage (e.g., > 20 V) to flash cells for a long time (e.g., > 3.5 ms), which degrades cell endurance and potentially delays user I/O requests. While a large body of prior work has proposed various techniques to mitigate the negative impact of erase operations, no work has yet investigated how erase latency should be set to fully exploit the potential of NAND flash memory; most existing techniques use a fixed latency for every erase operation which is set to cover the worst-case operating conditions. To address this, we propose AERO (Adaptive ERase Operation), a new erase scheme that dynamically adjusts erase latency to be just long enough for reliably erasing target cells, depending on the cells' current erase characteristics. AERO accurately predicts such near-optimal erase latency based on the number of fail bits during an erase operation. To maximize its benefits, we further optimize AERO in two aspects. First, at the beginning of an erase operation, AERO attempts to erase the ce

A fast and scalable scheme for multi-qubit resetting in superconducting quantum processors is proposed by exploiting the feasibility of frequency-tunable transmon qubits and transmon-like couplers to engineer a full programmable superconducting erasing head. The scalability of the device is verified by simultaneously resetting two qubits, where we show that collectivity effects may emerge as an fundamental ingredient to speed up the erasing process. Conversely, we also describe the appearance of decoherence-free subspace in multi-qubit chips, causing it to damage the device performance. To overcome this problem, a special set of parameters for the tunable frequency coupler is proposed, which allows us to erase even states within such subspace. To end, we offer a proposal to buildup integrated superconducting processors that can be efficiently connected to erasure heads in a scalable way.

The goal of weakly supervised video anomaly detection is to learn a detection model using only video-level labeled data. However, prior studies typically divide videos into fixed-length segments without considering the complexity or duration of anomalies. Moreover, these studies usually just detect the most abnormal segments, potentially overlooking the completeness of anomalies. To address these limitations, we propose a Dynamic Erasing Network (DE-Net) for weakly supervised video anomaly detection, which learns multi-scale temporal features. Specifically, to handle duration variations of abnormal events, we first propose a multi-scale temporal modeling module, capable of extracting features from segments of varying lengths and capturing both local and global visual information across different temporal scales. Then, we design a dynamic erasing strategy, which dynamically assesses the completeness of the detected anomalies and erases prominent abnormal segments in order to encourage the model to discover gentle abnormal segments in a video. The proposed method obtains favorable performance compared to several state-of-the-art approaches on three datasets: XD-Violence, TAD, and UCF

Concept erasure in text-to-image diffusion models aims to disable pre-trained diffusion models from generating images related to a target concept. To perform reliable concept erasure, the properties of robustness and locality are desirable. The former refrains the model from producing images associated with the target concept for any paraphrased or learned prompts, while the latter preserves its ability in generating images with non-target concepts. In this paper, we propose Reliable Concept Erasing via Lightweight Erasers (Receler). It learns a lightweight Eraser to perform concept erasing while satisfying the above desirable properties through the proposed concept-localized regularization and adversarial prompt learning scheme. Experiments with various concepts verify the superiority of Receler over previous methods.

Text-to-Image models such as Stable Diffusion have shown impressive image generation synthesis, thanks to the utilization of large-scale datasets. However, these datasets may contain sexually explicit, copyrighted, or undesirable content, which allows the model to directly generate them. Given that retraining these large models on individual concept deletion requests is infeasible, fine-tuning algorithms have been developed to tackle concept erasing in diffusion models. While these algorithms yield good concept erasure, they all present one of the following issues: 1) the corrupted feature space yields synthesis of disintegrated objects, 2) the initially synthesized content undergoes a divergence in both spatial structure and semantics in the generated images, and 3) sub-optimal training updates heighten the model's susceptibility to utility harm. These issues severely degrade the original utility of generative models. In this work, we present a new approach that solves all of these challenges. We take inspiration from the concept of classifier guidance and propose a surgical update on the classifier guidance term while constraining the drift of the unconditional score term. Furthe

We introduce partial loop-erasing operators. We show that by applying a refinement sequence of partial loop-erasing operators to a finite Markov chain, we get a process equivalent to the chronological loop-erased Markov chain. As an application, we construct loop-erased random paths on bounded domains of resistance spaces as the weak limit of the loop erasure of the Markov chains on a sequence of finite sets approximating the space, and the limit is independent of the approximating sequences. The random paths we constructed are simple paths almost surely, and they can be viewed as the loop-erasure of the paths of the diffusion process. Finally, we show that the scaling limit of the loop-erased random walks on the Sierpiński carpet graphs exists, and is equivalent to the loop-erased random paths on the Sierpińksi carpet.

The famous theorem by Chomsky and Schützenberger (CST) says that every context-free language $L$ over an alphabet $Σ$ is representable as $h(D \cap R)$, where $D$ is a Dyck language over a set $Ω$ of brackets, $R$ is a local language and $h$ is an alphabetic homomorphism that erases unboundedly many symbols. Berstel found that the number of erasures can be linearly limited if the grammar is in Greibach normal form; Berstel and Boasson (and later, independently, Okhotin) proved a non-erasing variant of CST for grammars in Double Greibach Normal Form. In all these CST statements, however, the size of the Dyck alphabet $Ω$ depends on the grammar size for $L$. In the Stanley variant of the CST, $|Ω|$ only depends on $|Σ|$ and not on the grammar, but the homomorphism erases many more symbols than in the other versions of CST; also, the regular language $R$ is strictly locally testable but not local. We prove a new version of CST which combines both features of being non-erasing and of using a grammar-independent alphabet. In our construction, $|Ω|$ is polynomial in $|Σ|$, namely $O(|Σ|^{46})$, and the regular language $R$ is strictly locally testable. Using a recent generalization of Me

The quantum eraser variant of the double-slit experiment, and its 'delayed choice' sub-variant, are considered from the perspective of weak value and weak measurement theory (which is briefly reintroduced here). The interference fringes that appear when measuring certain spin states, which can then be 'erased' when measuring other spin states, are shown to be anomalous weak values that depend on particular post-selection choices. By framing the choice of spin measurement as a weak value of a certain weak measurement, it is then made clear what physical claims can and cannot be made about what occurs in the quantum eraser experiment. Specifically, claims about the choice of spin-state `retrocausally' influencing the choice of slit(s) for the particles to travel through are discredited, and a simple framework is presented for understanding how the fringes arise and why they can be 'erased'.

Considering the delayed-choice quantum eraser using a Mach-Zehnder interferometer with a nonsymmetric beam splitter, we explicitly demonstrate that it shares exactly the same formal structure with the Einstein-Podolsky-Rosen-Bohm (EPR-Bohm) experiment. Therefore, the effect of quantum erasure can be understood in terms of the standard EPR correlation. Nevertheless, the quantum eraser still raises a conceptual issue beyond the standard EPR paradox, if counterfactual reasoning is taken into account. Furthermore, the quantum eraser experiments can be classified into two major categories: the entanglement quantum eraser and the Scully-Drühl-type quantum eraser. These two types are formally equivalent to each other, but conceptually the latter presents a "mystery" more prominent than the former. In the Scully-Drühl-type quantum eraser, the statement that the which-way information can be influenced by the delayed-choice measurement is not purely a consequence of counterfactual reasoning but bears some factual significance. Accordingly, it makes good sense to say that the "record" of the which-way information is "erased" if the potentiality to yield a conclusive outcome that discriminates