{"id":"49e6c536-d713-4876-9c7c-71b8a10496b2","arxiv_id":"2605.30452","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A unifying framework for gradient aggregation in multi-objective optimization establishes convergence to Pareto stationarity under a sufficient alignment condition and introduces capped MGDA for robustness in adversarial federated learning.","lead":"This paper creates a single framework that explains how different multi-objective optimization methods combine gradients from competing objectives in machine learning. A smart generalist might read it to understand shared convergence properties across existing algorithms and how new variants can be designed for settings like federated learning.","discovery_kind":"unification","skeptic_critique":{"model":"grok-4.3","headline":"Dual-cone projection may exit the convex hull, breaking the sufficient condition for convergence","rationale":"The reader's weakest assumption is exactly the point at which the argument is least secure: whether dual-cone projection can be applied without invalidating the hull membership required by the convergence theorem. The abstract presents the projection as a broadening device but supplies no lemma or remark confirming preservation of the hull, making this the single load-bearing gap.","tokens_in":1713,"tokens_out":345,"duration_ms":24596,"concrete_test":"Take two gradients g1, g2 in R^2 whose convex hull is the line segment between them. Choose an interior point d in conv{g1,g2} whose inner products with g1 and g2 have mixed signs. Compute its Euclidean projection onto the dual cone {v | <v,gi> <= 0, i=1,2}. Check whether the result lies inside the segment; repeat for several angle configurations between g1 and g2. If any projected vector exits the segment, the theorem does not cover the projected iterates.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central theorem requires a direction to be simultaneously non-conflicting and inside conv{g1,...,gm} to guarantee descent to Pareto stationarity. The paper uses dual-cone projection to enforce the non-conflicting property for aggregation rules that would otherwise violate it. Nothing in the abstract or claim description shows that the projected vector remains inside the convex hull; if the projection can map a point in conv{G} to a point outside conv{G}, the hypothesis of the convergence theorem is no longer satisfied and the derived rates do not apply.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes a unified framework for gradient aggregation in multi-objective optimization. It introduces a sufficient alignment condition leading to a theorem that non-conflicting directions chosen within the convex hull of the gradients ensure convergence to Pareto stationarity, from which optimal rates are derived. The framework allows enforcing the condition via dual-cone projection, provides a primal-dual perspective unifying existing algorithms, introduces capped MGDA based on CVaR, and includes experimental validation on synthetic and practical tasks like adversarial federated learning.","tokens_in":1795,"tokens_out":358,"duration_ms":18197,"significance":"If the central theorem applies after projection, the work offers a general analysis tool for MOO methods, potentially clarifying their relationships and enabling new convergent variants. The primal perspective and new algorithm are positive contributions.","major_comments":[{"comment":"The central theorem requires a direction to be simultaneously non-conflicting and inside conv{g1,...,gm} to guarantee descent to Pareto stationarity. The paper uses dual-cone projection to enforce the non-conflicting property. Nothing shows that the projected vector remains inside the convex hull; if the projection can map a point in conv{G} to a point outside conv{G}, the hypothesis of the convergence theorem is no longer satisfied and the derived rates do not apply. This is load-bearing for all convergence claims.","section":"Central theorem (as stated in abstract and developed in the main analysis)"}],"minor_comments":[{"comment":"The abstract mentions experiments on synthetic problems and practical benchmarks, but additional details on statistical significance, hyperparameter sensitivity, and comparison metrics would improve clarity.","section":"Experiments"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their careful reading of the manuscript and for identifying this important point regarding the central theorem. We address the comment below.","responses":[{"response":"We agree that the manuscript does not explicitly prove that the dual-cone projection of a vector from conv{G} necessarily remains inside conv{G}. This is a valid observation, and the current presentation leaves a gap in rigorously connecting the projection step to the hypothesis of the convergence theorem. In the revised manuscript we will add a lemma establishing the required invariance (either by showing that the particular projection operator employed maps conv{G} into itself, or by redefining the feasible set as the intersection of the dual cone with conv{G} and proving that the resulting projection satisfies both conditions simultaneously). This will be accompanied by the corresponding updates to the statement of the main theorem and the derived rates.","revision_made":"yes","referee_comment":"[Central theorem (as stated in abstract and developed in the main analysis)] The central theorem requires a direction to be simultaneously non-conflicting and inside conv{g1,...,gm} to guarantee descent to Pareto stationarity. The paper uses dual-cone projection to enforce the non-conflicting property. Nothing shows that the projected vector remains inside the convex hull; if the projection can map a point in conv{G} to a point outside conv{G}, the hypothesis of the convergence theorem is no longer satisfied and the derived rates do not apply. This is load-bearing for all convergence claims."}],"tokens_in":1298,"tokens_out":322,"duration_ms":22696,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper's core is a unifying framework for how MOO methods combine gradients, built around a sufficient alignment condition. Non-conflicting directions inside the convex hull of the component gradients are enough to guarantee convergence to Pareto stationarity, and optimal rates follow from that. They add a primal optimization view that organizes existing algorithms and lets them derive new ones, including capped MGDA from a CVaR formulation, which they apply to adversarial federated learning.\n\nIt does a solid job pulling prior methods together and showing their relationships through the primal-dual lens. The dual-cone projection idea to make more aggregation rules feasible is a reasonable way to widen the set of methods that get guarantees. Experiments on synthetic cases and practical benchmarks are the usual check.\n\nThe soft spot is the one flagged in the stress test. The theorem requires the direction to be both non-conflicting and inside the convex hull. Dual-cone projection enforces the first property, but the abstract gives no argument that the result remains in the hull. If the projection can move a point outside, the convergence claim does not apply to those cases. The paper says the projection broadens the scope, so the full text must show either that it preserves the hull or that another mechanism keeps the condition intact. This is the part that needs the clearest proof.\n\nThis work is for people already working on multi-objective optimization in machine learning, especially those dealing with trade-offs in training or federated settings. A reader who wants a cleaner way to compare aggregation rules or to design new ones would get something from it. The unification and the new algorithm are substantive enough that the paper deserves a serious referee rather than a desk reject, though the projection step will probably need tightening.","headline":"The paper unifies MOO gradient aggregation under an alignment condition for convergence rates and adds a capped MGDA variant, but the dual-cone projection step needs explicit confirmation that it stays inside the convex hull.","tokens_in":2290,"tokens_out":433,"would_cite":false,"duration_ms":23027,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Non-conflicting directions within the convex hull of gradients ensure convergence to Pareto stationarity in multi-objective optimization.","keywords":["multi-objective optimization","gradient aggregation","Pareto stationarity","convergence rates","MGDA","dual cone projection","federated learning"],"falsifier":"Run a simple two-objective quadratic problem where the convex hull contains non-conflicting directions; an aggregation rule that systematically outputs conflicting directions should fail to reach Pareto stationarity while a projected rule succeeds.","tokens_in":2586,"feed_emoji":"📐","tokens_out":636,"duration_ms":18206,"temperature":0.7,"pith_summary":"The paper develops a unifying framework for gradient aggregation across multiple objectives. It proves that directions staying inside the convex hull and free of mutual conflict form a sufficient condition for convergence to Pareto stationarity. From this single condition the authors obtain optimal convergence rates and show how a dual-cone projection step makes the condition feasible for a wider set of rules. The same lens recovers existing algorithms through a primal optimization view and produces a new capped MGDA variant based on CVaR. The result matters for any setting where several loss surfaces must be balanced simultaneously, because it replaces case-by-case proofs with one reusable argument.","feed_headline":"Non-conflicting gradients converge to Pareto stationarity","feed_subtitle":"Directions inside the convex hull of component gradients yield optimal rates under one reusable alignment condition.","key_machinery":"The sufficient alignment condition on non-conflicting directions chosen inside the convex hull of the component gradients, which directly yields convergence to Pareto stationarity.","core_discovery":"Non-conflicting directions, when chosen within the convex hull of gradients, form a fundamental sufficient condition for convergence to Pareto stationarity, from which optimal rates are derived. Feasibility is ensured through projection onto the dual cone, and a primal optimization perspective unifies established algorithms while enabling new variants such as capped MGDA.","pith_inferences":["Aggregation rules that occasionally violate alignment may still converge in practice but lose the paper's rate guarantees.","The framework suggests testing whether the dual-cone projection improves stability on non-convex or stochastic multi-objective problems.","Similar alignment ideas could be applied to federated or distributed settings beyond the adversarial case shown."],"forward_implications":["Any aggregation rule whose output satisfies the alignment condition inherits the optimal convergence rates to Pareto stationarity.","Projection onto the dual cone extends convergence guarantees to a broader family of methods that would otherwise produce conflicting directions.","Capped MGDA, derived from the CVaR formulation, inherits the same rates and shows improved robustness in adversarial federated learning.","The primal optimization view recovers MGDA, linear scalarization, and other standard methods as special cases with explicit relationships among them."],"fun_headline_variants":["Convex hull of gradients yields Pareto stationarity","Non-conflicting directions ensure MOO convergence","Dual cone projection broadens MOO convergence guarantees","Primal view unifies gradient aggregation methods"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The sufficient alignment condition holds for the directions produced by the aggregation rule or can be enforced via dual-cone projection without invalidating the convergence argument.","fun_headline_variants_meta":{"raw":{"variants":["Convex hull of gradients yields Pareto stationarity","Non-conflicting directions ensure MOO convergence","Dual cone projection broadens MOO convergence guarantees","Primal view unifies gradient aggregation methods"]},"model":"grok-4.3","cost_usd":0.009495,"raw_usage":{"total_tokens":4221,"prompt_tokens":631,"num_sources_used":0,"completion_tokens":54,"cost_in_usd_ticks":94949500,"prompt_tokens_details":{"text_tokens":631,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3536,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":631,"tokens_out":54,"duration_ms":28910,"temperature":1.0,"reasoning_tokens":3536,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-29T08:42:13.623940+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Run a simple two-objective quadratic problem where the convex hull contains non-conflicting directions; an aggregation rule that systematically outputs conflicting directions should fail to reach Pareto stationarity while a projected rule succeeds.","supporting_citations":[],"review_version":1}